diff --git a/src/__tests__/integration/server-error-contracts.test.ts b/src/__tests__/integration/server-error-contracts.test.ts
index 1fb674a7e..3138da183 100644
--- a/src/__tests__/integration/server-error-contracts.test.ts
+++ b/src/__tests__/integration/server-error-contracts.test.ts
@@ -1692,7 +1692,7 @@ describe("task errors", () => {
     )
   })
 
-  it("vault_create_task with invalid date", async () => {
+  it("vault_create_task with a date the calendar lacks", async () => {
     const result = await callTool({
       client,
       name: "vault_create_task",
@@ -1703,10 +1703,18 @@ describe("task errors", () => {
         due: "2026-02-30",
       },
     })
-    expectToolError(result, "invalid date")
+    expect(result).toEqual({
+      content: [
+        {
+          type: "text",
+          text: '[Error]: invalid date: due "2026-02-30" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+        },
+      ],
+      isError: true,
+    })
   })
 
-  it("vault_update_task with invalid date", async () => {
+  it("vault_update_task with a malformed date", async () => {
     const result = await callTool({
       client,
       name: "vault_update_task",
@@ -1716,7 +1724,10 @@ describe("task errors", () => {
         due: "not-a-date",
       },
     })
-    expectToolError(result, "invalid date")
+    expect(result).toEqual({
+      content: [{ type: "text", text: '[Error]: invalid date: due "not-a-date" (use YYYY-MM-DD)' }],
+      isError: true,
+    })
   })
 
   it("vault_update_task — cannot move a sub-task to a heading", async () => {
@@ -2535,3 +2546,75 @@ describe("tag must not be empty", () => {
     })
   })
 })
+
+// ── Date filter values ───────────────────────────────────────
+
+describe("date filter values", () => {
+  it("vault_search rejects a date filter not in YYYY-MM-DD form", async () => {
+    const result = await callTool({
+      client,
+      name: "vault_search",
+      args: { query: "note", filters: { modified: { after: "yesterday" } } },
+    })
+    expect(result).toEqual({
+      content: [
+        {
+          type: "text",
+          text: '[Error]: invalid modified.after date: "yesterday". Use YYYY-MM-DD (e.g. 2026-07-03).',
+        },
+      ],
+      isError: true,
+    })
+  })
+
+  it("vault_search rejects a date filter naming a day the calendar lacks", async () => {
+    const result = await callTool({
+      client,
+      name: "vault_search",
+      args: { query: "note", filters: { created: { on: "2026-02-30" } } },
+    })
+    expect(result).toEqual({
+      content: [
+        {
+          type: "text",
+          text: '[Error]: created.on date "2026-02-30" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+        },
+      ],
+      isError: true,
+    })
+  })
+
+  it("vault_list_tasks rejects a date filter not in YYYY-MM-DD form", async () => {
+    const result = await callTool({
+      client,
+      name: "vault_list_tasks",
+      args: { due: { before: "July 3rd" } },
+    })
+    expect(result).toEqual({
+      content: [
+        {
+          type: "text",
+          text: '[Error]: invalid due.before date: "July 3rd". Use YYYY-MM-DD (e.g. 2026-07-03).',
+        },
+      ],
+      isError: true,
+    })
+  })
+
+  it("vault_list_tasks rejects a date filter naming a day the calendar lacks", async () => {
+    const result = await callTool({
+      client,
+      name: "vault_list_tasks",
+      args: { done: { after: "2026-04-31" } },
+    })
+    expect(result).toEqual({
+      content: [
+        {
+          type: "text",
+          text: '[Error]: done.after date "2026-04-31" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+        },
+      ],
+      isError: true,
+    })
+  })
+})
diff --git a/src/utils/__tests__/classify-iso-date.test.ts b/src/utils/__tests__/classify-iso-date.test.ts
new file mode 100644
index 000000000..216207490
--- /dev/null
+++ b/src/utils/__tests__/classify-iso-date.test.ts
@@ -0,0 +1,33 @@
+import { describe, expect, it } from "vitest"
+import { classifyIsoDate } from "../classify-iso-date.js"
+
+describe("classifyIsoDate", () => {
+  it.each([
+    { label: "an ordinary day", text: "2026-07-03" },
+    { label: "February 29 in a leap year", text: "2024-02-29" },
+  ])("classifies $label as a calendar date", ({ text }) => {
+    expect(classifyIsoDate(text)).toBe("calendar-date")
+  })
+
+  it.each([
+    { label: "February 30", text: "2026-02-30" },
+    { label: "February 29 outside a leap year", text: "2025-02-29" },
+    { label: "month 13", text: "2026-13-01" },
+    { label: "month 00", text: "2026-00-10" },
+    { label: "day 00", text: "2026-02-00" },
+  ])("classifies $label as not a calendar date", ({ text }) => {
+    expect(classifyIsoDate(text)).toBe("not-a-calendar-date")
+  })
+
+  it.each([
+    { label: "a single-digit month and day", text: "2026-7-3" },
+    { label: "a timestamp", text: "2026-07-03T10:30:00" },
+    { label: "a five-digit year", text: "20260-01-01" },
+    { label: "slashes", text: "2026/07/03" },
+    { label: "a month name", text: "March 10" },
+    { label: "surrounding whitespace", text: " 2026-07-03" },
+    { label: "an empty string", text: "" },
+  ])("classifies $label as malformed", ({ text }) => {
+    expect(classifyIsoDate(text)).toBe("malformed")
+  })
+})
diff --git a/src/utils/classify-iso-date.ts b/src/utils/classify-iso-date.ts
new file mode 100644
index 000000000..5576c79a7
--- /dev/null
+++ b/src/utils/classify-iso-date.ts
@@ -0,0 +1,21 @@
+import { DateTime } from "luxon"
+
+type IsoDateClass = "calendar-date" | "malformed" | "not-a-calendar-date"
+
+/** Classifies text by whether it names a real day in YYYY-MM-DD form. The two failures:
+ *  - "malformed": not in that form, such as "2026-7-3", "March 10" or a timestamp
+ *  - "not-a-calendar-date": in that form, but the month or day doesn't exist,
+ *    such as 2026-02-30 or 2026-13-01
+ *  They stay apart so an error message can name the failure: a client told only
+ *  "use YYYY-MM-DD" sends 2026-02-30 again, since it already used that form. */
+export const classifyIsoDate = (text: string): IsoDateClass => {
+  const parsedDate = DateTime.fromFormat(text, "yyyy-MM-dd")
+
+  if (parsedDate.isValid) return "calendar-date"
+
+  // Luxon checks the form first ("unparsable" when it doesn't match), then the
+  // calendar ("unit out of range" for a month or day that doesn't exist)
+  if (parsedDate.invalidReason === "unit out of range") return "not-a-calendar-date"
+
+  return "malformed"
+}
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/default.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/default.json
index 0d0a8e74c..5e88da5a6 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/default.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/default.json
@@ -650,7 +650,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1283,7 +1283,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/disabled-tools.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/disabled-tools.json
index fd7aad709..62bb8a65e 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/disabled-tools.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/disabled-tools.json
@@ -652,7 +652,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1225,7 +1225,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/embedding-off.json
index dcd5aea60..37483f114 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/embedding-off.json
@@ -652,7 +652,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1285,7 +1285,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off+embedding-off.json
index 4e1ccd4c7..9eecaa160 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off+embedding-off.json
@@ -612,7 +612,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1201,7 +1201,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off.json
index 5c80dcbb1..1dcb44a24 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/file-tools-off.json
@@ -611,7 +611,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1200,7 +1200,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+embedding-off.json
index a82325bc9..05e75f098 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+embedding-off.json
@@ -553,7 +553,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1146,7 +1146,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off+embedding-off.json
index f6e22f5ab..2aa784b1a 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off+embedding-off.json
@@ -513,7 +513,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1062,7 +1062,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off.json
index 02e08dd0f..6a8c9930e 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off+file-tools-off.json
@@ -512,7 +512,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1061,7 +1061,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off.json
index a1f2c2e51..72ddaafa6 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/memory-off.json
@@ -552,7 +552,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1145,7 +1145,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/obsidian-sync.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/obsidian-sync.json
index bf47bbb6f..3f68d1ebd 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/obsidian-sync.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/obsidian-sync.json
@@ -652,7 +652,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -1285,7 +1285,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+embedding-off.json
index fe714cdb4..32ad91652 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+embedding-off.json
@@ -334,7 +334,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -774,7 +774,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off+embedding-off.json
index 98246ac3d..97b66284e 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off+embedding-off.json
@@ -294,7 +294,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -690,7 +690,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off.json
index d892adf63..732c3a85d 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+file-tools-off.json
@@ -293,7 +293,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -689,7 +689,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+embedding-off.json
index 4d8ab402b..26a839037 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+embedding-off.json
@@ -281,7 +281,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -681,7 +681,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off+embedding-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off+embedding-off.json
index 1d8dacbba..38b9c5780 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off+embedding-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off+embedding-off.json
@@ -241,7 +241,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -597,7 +597,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"deployment\", filters: { properties: { status: \"active\" } } })\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"fts\" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off.json
index 1ae2d50fe..7e53a6f12 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off+file-tools-off.json
@@ -240,7 +240,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -596,7 +596,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off.json
index 4abe2fff2..86fd210c0 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly+memory-off.json
@@ -280,7 +280,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -680,7 +680,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly.json b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly.json
index 117ef35f3..ede14a24c 100644
--- a/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly.json
+++ b/src/vault-mcp/mcp-core/__tests__/__snapshots__/tool-surface/readonly.json
@@ -333,7 +333,7 @@
     {
       "name": "vault_list_tasks",
       "title": "List Tasks",
-      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- A malformed or calendar-invalid date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
+      "description": "List checkbox tasks across the whole vault with structured filters — the Tasks-plugin data model over MCP. Both task metadata formats are indexed: emoji signifiers (📅 due, ⏳ scheduled, 🛫 start, ➕ created, ✅ done, ❌ cancelled, 🔺⏫🔼🔽⏬ priority, 🔁 recurrence, 🏁 onCompletion, 🆔/⛔ dependencies) and Dataview inline fields ([due:: 2026-07-04], [priority:: high], ...). Every result carries its attribution — note path, folder, line number, and the nearest heading when the task sits under one (the lane on a Kanban board) — so no follow-up reads are needed to locate a task. Task lines inside fenced code blocks and %% %% comment blocks are not indexed. Checkboxes typed NON_TASK in the Tasks plugin's status registry are also excluded.\n\nExample: vault_list_tasks({ due: { before: \"2026-07-04\" } }) — overdue triage; the default status (not_done) and sort (due ascending) make this the \"what's overdue?\" call\nExample: vault_list_tasks({ path: \"Code Projects/vault-cortex/TASKS.md\", heading: [\"Active\", \"Up Next\", \"Waiting On\"], sort_by: \"position\" }) — actionable Kanban lanes in board order\nExample: vault_list_tasks({ folder: \"Code Projects/vault-cortex\" }) — all open tasks across a project tree (TASKS.md + task-notes/ subdirectories)\nExample: vault_list_tasks({ status: \"done\", done: { after: \"2026-06-26\" } }) — what got completed this week\nExample: vault_list_tasks({ top_level_only: true, path: \"TASKS.md\" }) — board cards only, excluding checklist sub-items\n\nWhen to use: Reading task status and order on one board (path + sort_by: \"position\"; add status: \"all\" to include done and cancelled cards). Also any vault-wide task triage question — \"what's overdue?\", \"what's open per project?\", \"what did I finish this week?\" — in one call instead of per-board reads.\nPrefer vault_read_note (heading mode) only when you need a lane's verbatim Markdown or a task's state right after a write. Prefer vault_search for full-text queries over note content.\n\nBehavior: Reads the search index, which picks up a file change within a few seconds, so a task written moments ago may still show its old state.\n\nParameters:\n- status: virtual values expand in arrays — [\"not_done\", \"done\"] matches todo + in_progress + done.\n- due / scheduled / start / done / created / cancelled: a date filter only matches tasks that HAVE that date.\n- folder: a whole folder, subfolders included (\"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"), ignoring ASCII letter case.\n\nErrors:\n- \"invalid due.before date: …\" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- \"path must end in .md\" — add the \".md\" extension\n- tag must not be empty after its leading \"#\" — tag is \"#\" alone; pass the tag name\n- No matches returns { total: 0, tasks: [] }, not an error\n\nReturns: JSON { total, tasks }. Every task carries path, line, status, status_char, description, folder, depth (0 for top-level, 1+ for sub-tasks), is_kanban_task, depends_on, and tags (the arrays are [] when empty). Every other field appears only when the task has it: heading (nearest heading above the task), created/scheduled/start/due/done/cancelled dates, priority, recurrence, on_completion, task_id, block_id, parent_block_id (sub-tasks whose parent carries a ^block-id), done_lanes (Kanban boards only), and subtask_progress — { done, total } over the task's DIRECT checklist children, present only when the task has a checklist; done counts status \"done\" only (a cancelled child counts toward total, not done), and the counts ignore the query's filters — so a filtered or top_level_only read still shows each card's checklist progress.",
       "inputSchema": {
         "type": "object",
         "properties": {
@@ -773,7 +773,7 @@
     {
       "name": "vault_search",
       "title": "Search Notes",
-      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- A malformed or calendar-invalid created/modified date filter throws with remediation text (\"Use YYYY-MM-DD\")\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
+      "description": "Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — \"career aspirations\" finds notes about \"goals\" and \"targets\". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the \"narrow by metadata, search by text\" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.\n\nFilters — all conditions AND-combine with each other and the text query:\n- folder: a whole folder, subfolders included — \"Projects\" covers \"Projects/Archive\" but not \"ProjectsOld/\"; ignores ASCII letter case\n- tags: every listed frontmatter tag must match, in any letter case; \"project\" also matches \"project/a\"\n- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: \"active\" }). Values compare by exact type — pass a number as a number, not \"4\". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.\n- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter\n\nExample: vault_search({ query: \"kubernetes networking\", filters: { tags: [\"reference\"] } })\nExample: vault_search({ query: \"meeting notes\", filters: { type: \"meeting\", folder: \"Work\" } })\nExample: vault_search({ query: \"decision\", filters: { modified: { after: \"2026-06-30\" } } }) — matching notes touched in July or later\nExample: vault_search({ query: \"how the server watches for file changes\" }) — semantic: finds notes about chokidar and file watchers even without those exact terms\n\nWhen to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.\nPrefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.\n\nErrors:\n- No matches returns { results: [], total: 0 }, not an error\n- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error\n- \"invalid created.on date: …\" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form\n- \"… is not a calendar date\" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date\n- tag must not be empty after its leading \"#\" — a tags entry is only \"#\"; pass the tag name\n\nReturns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode (\"hybrid\" or \"fts\"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — \"hybrid\" when vector embeddings contributed, \"fts\" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is \"note\" for markdown notes or \"file\" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. \".canvas\", \".pdf\", \".txt\"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.",
       "inputSchema": {
         "type": "object",
         "properties": {
diff --git a/src/vault-mcp/mcp-core/tools/search-tools.ts b/src/vault-mcp/mcp-core/tools/search-tools.ts
index 526db335c..5dce95936 100644
--- a/src/vault-mcp/mcp-core/tools/search-tools.ts
+++ b/src/vault-mcp/mcp-core/tools/search-tools.ts
@@ -17,6 +17,21 @@ export const registerSearchTools = ({
   logger: sessionLogger,
   config,
 }: ToolRegistrationContext): void => {
+  // Filters and errors are the same with or without embeddings, so both
+  // vault_search description variants share these two sections
+  const searchFiltersSection = `Filters — all conditions AND-combine with each other and the text query:
+- folder: a whole folder, subfolders included — "Projects" covers "Projects/Archive" but not "ProjectsOld/"; ignores ASCII letter case
+- tags: every listed frontmatter tag must match, in any letter case; "project" also matches "project/a"
+- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: "active" }). Values compare by exact type — pass a number as a number, not "4". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.
+- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter`
+
+  const searchErrorsSection = `Errors:
+- No matches returns { results: [], total: 0 }, not an error
+- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error
+- "invalid created.on date: …" (any created/modified bound) — the bound isn't in YYYY-MM-DD form; use that form
+- "… is not a calendar date" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date
+- tag must not be empty after its leading "#" — a tags entry is only "#"; pass the tag name`
+
   registerTool(
     TOOL_NAMES.VAULT_SEARCH,
     {
@@ -24,11 +39,7 @@ export const registerSearchTools = ({
       description: config.embeddingEnabled
         ? `Hybrid search across all vault notes, ranked by combined keyword and semantic relevance using Reciprocal Rank Fusion (RRF) — combining FTS5 keyword matching with vector similarity. Results are refined by a cross-encoder reranker using position-aware score blending when available. Semantic matching finds notes even when exact keywords differ — "career aspirations" finds notes about "goals" and "targets". Falls back to keyword-only (FTS5 BM25) transparently while embeddings are being built. Combine a text query with structured filters to narrow results by metadata — the "narrow by metadata, search by text" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.
 
-Filters — all conditions AND-combine with each other and the text query:
-- folder: a whole folder, subfolders included — "Projects" covers "Projects/Archive" but not "ProjectsOld/"; ignores ASCII letter case
-- tags: every listed frontmatter tag must match, in any letter case; "project" also matches "project/a"
-- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: "active" }). Values compare by exact type — pass a number as a number, not "4". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.
-- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter
+${searchFiltersSection}
 
 Example: vault_search({ query: "kubernetes networking", filters: { tags: ["reference"] } })
 Example: vault_search({ query: "meeting notes", filters: { type: "meeting", folder: "Work" } })
@@ -38,20 +49,12 @@ Example: vault_search({ query: "how the server watches for file changes" }) —
 When to use: The primary discovery tool for content-based queries, optionally constrained by metadata. Semantic matching bridges vocabulary gaps — try natural-language queries, not just keywords.
 Prefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.
 
-Errors:
-- No matches returns { results: [], total: 0 }, not an error
-- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error
-- A malformed or calendar-invalid created/modified date filter throws with remediation text ("Use YYYY-MM-DD")
-- tag must not be empty after its leading "#" — a tags entry is only "#"; pass the tag name
+${searchErrorsSection}
 
 Returns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode ("hybrid" or "fts"), and reranked (boolean — true when cross-encoder reranking refined the ordering). search_mode indicates which ranking was used — "hybrid" when vector embeddings contributed, "fts" when only keyword matching was available. score reflects combined relevance (higher = more relevant). kind is "note" for markdown notes or "file" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. ".canvas", ".pdf", ".txt"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.`
         : `Full-text search across all vault notes, ranked by relevance. Combine a text query with structured filters to narrow results by metadata — the "narrow by metadata, search by text" pattern. Unquoted terms use implicit AND with porter stemming; wrap in double quotes for exact phrases; punctuated terms (vault-cortex, deploy/local) are matched as exact adjacent-word phrases automatically.
 
-Filters — all conditions AND-combine with each other and the text query:
-- folder: a whole folder, subfolders included — "Projects" covers "Projects/Archive" but not "ProjectsOld/"; ignores ASCII letter case
-- tags: every listed frontmatter tag must match, in any letter case; "project" also matches "project/a"
-- properties: arbitrary frontmatter key-value pairs, supports string/number/boolean (e.g. { status: "active" }). Values compare by exact type — pass a number as a number, not "4". Exception: checkbox values are stored as 1 and 0, so pass true to match 1 and false to match 0. A list property matches when any element equals the value.
-- created / modified: bounds compare whole calendar days, server-local, so on matches anywhere within the day. Notes without a parseable created property never match a created filter
+${searchFiltersSection}
 
 Example: vault_search({ query: "kubernetes networking", filters: { tags: ["reference"] } })
 Example: vault_search({ query: "meeting notes", filters: { type: "meeting", folder: "Work" } })
@@ -61,11 +64,7 @@ Example: vault_search({ query: "deployment", filters: { properties: { status: "a
 When to use: The primary discovery tool for content-based queries, optionally constrained by metadata.
 Prefer vault_search_by_tag for tag-only queries without text. Prefer vault_search_by_folder for browsing a folder. Prefer vault_search_by_property for metadata-only queries. Prefer vault_recent_notes for time-based browsing.
 
-Errors:
-- No matches returns { results: [], total: 0 }, not an error
-- Malformed query syntax is sanitized automatically — the tool never throws a query syntax error
-- A malformed or calendar-invalid created/modified date filter throws with remediation text ("Use YYYY-MM-DD")
-- tag must not be empty after its leading "#" — a tags entry is only "#"; pass the tag name
+${searchErrorsSection}
 
 Returns: JSON with results array (path, title, snippet, score, tags, folder, type, kind, extension, created, modified, bytes), total (results returned, not all matches), search_mode ("fts" — keyword-only ranking), and reranked (always false in keyword-only mode). kind is "note" for markdown notes or "file" for non-markdown content (canvas, PDF, and text files — .txt, .csv, .json, .xml, .svg, .log, .yaml, .yml, .base); file results also carry extension (e.g. ".canvas", ".pdf", ".txt"). created is omitted when null. bytes is the on-disk file size. With include_leading_callout, each result also carries leading_callout ({ type, title, body }) when present.`,
       inputSchema: {
diff --git a/src/vault-mcp/mcp-core/tools/task-tools.ts b/src/vault-mcp/mcp-core/tools/task-tools.ts
index 3a7a21d35..474b69f46 100644
--- a/src/vault-mcp/mcp-core/tools/task-tools.ts
+++ b/src/vault-mcp/mcp-core/tools/task-tools.ts
@@ -43,7 +43,8 @@ Parameters:
 - folder: a whole folder, subfolders included ("Projects" covers "Projects/Archive" but not "ProjectsOld/"), ignoring ASCII letter case.
 
 Errors:
-- A malformed or calendar-invalid date filter throws with remediation text ("Use YYYY-MM-DD")
+- "invalid due.before date: …" (any date filter's bound) — the bound isn't in YYYY-MM-DD form; use that form
+- "… is not a calendar date" — the bound has the YYYY-MM-DD form but names a day that doesn't exist, such as 2026-02-30; pass a real date
 - "path must end in .md" — add the ".md" extension
 - tag must not be empty after its leading "#" — tag is "#" alone; pass the tag name
 - No matches returns { total: 0, tasks: [] }, not an error
diff --git a/src/vault-mcp/search/__tests__/hybrid-search.test.ts b/src/vault-mcp/search/__tests__/hybrid-search.test.ts
index f72fa3fae..696d2d393 100644
--- a/src/vault-mcp/search/__tests__/hybrid-search.test.ts
+++ b/src/vault-mcp/search/__tests__/hybrid-search.test.ts
@@ -624,7 +624,9 @@ Content about quarterly planning and roadmaps.
           { query: "anything", filters: { modified: { on: "bad" } } },
           logger,
         ),
-      ).rejects.toThrow('invalid modified.on date: "bad". Use YYYY-MM-DD (e.g. 2026-07-03).')
+      ).rejects.toThrow(
+        new Error('invalid modified.on date: "bad". Use YYYY-MM-DD (e.g. 2026-07-03).'),
+      )
     })
   })
 
diff --git a/src/vault-mcp/search/__tests__/search-index.test.ts b/src/vault-mcp/search/__tests__/search-index.test.ts
index 8ef9d2db2..603186e20 100644
--- a/src/vault-mcp/search/__tests__/search-index.test.ts
+++ b/src/vault-mcp/search/__tests__/search-index.test.ts
@@ -6680,7 +6680,7 @@ Shared datefilter content for boundary tests.
     }).toThrow(/^invalid created\.on date: "March 10"\. Use YYYY-MM-DD \(e\.g\. 2026-07-03\)\.$/)
   })
 
-  it("rejects a calendar-invalid created date", () => {
+  it("rejects a created date the calendar lacks, saying so rather than naming the format", () => {
     const dateIndex = indexWithCreatedDates()
     expect(() => {
       dateIndex.fullTextSearch(
@@ -6691,7 +6691,19 @@ Shared datefilter content for boundary tests.
         logger,
       )
     }).toThrow(
-      /^invalid created\.before date: "2026-02-31"\. Use YYYY-MM-DD \(e\.g\. 2026-07-03\)\.$/,
+      /^created\.before date "2026-02-31" is not a calendar date\. Pass a real date in YYYY-MM-DD format\.$/,
+    )
+  })
+
+  it("rejects a created.after date the calendar lacks, naming that bound", () => {
+    const dateIndex = indexWithCreatedDates()
+    expect(() => {
+      dateIndex.fullTextSearch(
+        { query: "datefilter", filters: { created: { after: "2026-04-31" } } },
+        logger,
+      )
+    }).toThrow(
+      /^created\.after date "2026-04-31" is not a calendar date\. Pass a real date in YYYY-MM-DD format\.$/,
     )
   })
 })
@@ -6787,7 +6799,7 @@ Shared datefilter content for mtime boundary tests.
     )
   })
 
-  it("rejects a calendar-invalid modified date", () => {
+  it("rejects a modified date the calendar lacks, saying so rather than naming the format", () => {
     const dateIndex = indexWithModifiedTimes()
     expect(() => {
       dateIndex.fullTextSearch(
@@ -6798,7 +6810,7 @@ Shared datefilter content for mtime boundary tests.
         logger,
       )
     }).toThrow(
-      /^invalid modified\.before date: "2026-02-31"\. Use YYYY-MM-DD \(e\.g\. 2026-07-03\)\.$/,
+      /^modified\.before date "2026-02-31" is not a calendar date\. Pass a real date in YYYY-MM-DD format\.$/,
     )
   })
 
diff --git a/src/vault-mcp/search/__tests__/task-queries.test.ts b/src/vault-mcp/search/__tests__/task-queries.test.ts
index 91997ff3f..ee7728c7f 100644
--- a/src/vault-mcp/search/__tests__/task-queries.test.ts
+++ b/src/vault-mcp/search/__tests__/task-queries.test.ts
@@ -445,14 +445,16 @@ describe("listTasks date filters", () => {
   it("rejects a malformed date with remediation text", () => {
     const index = createTestIndex()
     expect(() => index.listTasks({ due: { before: "July 3rd" } }, logger)).toThrow(
-      'invalid due.before date: "July 3rd". Use YYYY-MM-DD (e.g. 2026-07-03).',
+      new Error('invalid due.before date: "July 3rd". Use YYYY-MM-DD (e.g. 2026-07-03).'),
     )
   })
 
-  it("rejects a calendar-invalid date", () => {
+  it("rejects a date the calendar lacks, saying so rather than naming the format", () => {
     const index = createTestIndex()
     expect(() => index.listTasks({ due: { on: "2026-02-31" } }, logger)).toThrow(
-      'invalid due.on date: "2026-02-31". Use YYYY-MM-DD (e.g. 2026-07-03).',
+      new Error(
+        'due.on date "2026-02-31" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+      ),
     )
   })
 })
@@ -660,7 +662,7 @@ describe("listTasks scope filters", () => {
   it("rejects a path without the .md extension", () => {
     const index = createTestIndex()
     expect(() => index.listTasks({ path: "Inbox/notes" }, logger)).toThrow(
-      'path must end in ".md" (received "Inbox/notes")',
+      new Error('path must end in ".md" (received "Inbox/notes")'),
     )
   })
 
diff --git a/src/vault-mcp/search/search-queries.ts b/src/vault-mcp/search/search-queries.ts
index 8bb1bc6b9..9423a409d 100644
--- a/src/vault-mcp/search/search-queries.ts
+++ b/src/vault-mcp/search/search-queries.ts
@@ -1,8 +1,8 @@
 // ── Query methods bound by createSearchIndex ─────────────────
 
 import type Database from "better-sqlite3"
-import { DateTime } from "luxon"
 import type { Logger } from "../../logger.js"
+import { classifyIsoDate } from "../../utils/classify-iso-date.js"
 import { describeError } from "../../utils/describe-error.js"
 import { compareByUtf8Bytes } from "../../utils/compare-utf8-bytes.js"
 import { assertPathHasExtension } from "../../utils/assert-path-has-extension.js"
@@ -140,18 +140,27 @@ type FileContentMetadataRow = {
   bytes: number
 }
 
-// ── Full-text search ───────────────────────────────────────────
+// ── Date filter validation ─────────────────────────────────────
 
-/** Rejects a malformed date filter bound with remediation text — shared by
- *  the note created/modified filters and the task date filters.
- *  `fromFormat` with `yyyy-MM-dd` pins both the format (no time component,
- *  no shorthand) and calendar correctness (2026-02-31 fails) in one call. */
+/** Rejects a date filter bound that isn't in YYYY-MM-DD form or names a day
+ *  the calendar lacks.
+ *  - Used by the note created/modified filters and the task date filters
+ *  - vault_search and vault_list_tasks quote both messages in their Errors sections */
 const assertFilterDate = (value: string, filterName: string): void => {
-  if (!DateTime.fromFormat(value, "yyyy-MM-dd").isValid) {
+  const dateClass = classifyIsoDate(value)
+
+  if (dateClass === "calendar-date") return
+
+  if (dateClass === "malformed") {
     throw new Error(`invalid ${filterName} date: "${value}". Use YYYY-MM-DD (e.g. 2026-07-03).`)
   }
+  throw new Error(
+    `${filterName} date "${value}" is not a calendar date. Pass a real date in YYYY-MM-DD format.`,
+  )
 }
 
+// ── Full-text search ───────────────────────────────────────────
+
 export const fullTextSearch = (
   context: SearchQueryContext,
   params: {
@@ -260,6 +269,8 @@ export const fullTextSearch = (
   if (params.filters?.modified) {
     const { on, before, after } = params.filters.modified
 
+    // assertFilterDate checks each bound before dayToEpochMsRange converts it,
+    // so a bad bound gets assertFilterDate's message, never the converter's
     if (on !== undefined) {
       assertFilterDate(on, "modified.on")
       const dayRange = dayToEpochMsRange(on)
@@ -996,8 +1007,15 @@ export const listTasks = (
     { column: "created", filter: params.created },
     { column: "cancelled", filter: params.cancelled },
   ]
+
+  // column does two jobs:
+  // - names the SQL column, safe to interpolate because its type allows only
+  //   the six names above
+  // - names the filter in error messages, which vault_list_tasks' Errors
+  //   section quotes ("invalid due.before date: …")
   for (const { column, filter } of dateFilters) {
-    if (filter === undefined) continue
+    if (!filter) continue
+
     if (filter.on !== undefined) {
       assertFilterDate(filter.on, `${column}.on`)
       conditions.push(`t.${column} = ?`)
diff --git a/src/vault-mcp/vault-operations/__tests__/task-mutations.test.ts b/src/vault-mcp/vault-operations/__tests__/task-mutations.test.ts
index 1f96bed37..8bfccd214 100644
--- a/src/vault-mcp/vault-operations/__tests__/task-mutations.test.ts
+++ b/src/vault-mcp/vault-operations/__tests__/task-mutations.test.ts
@@ -754,7 +754,7 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('note not found: "missing.md"')
+      ).rejects.toThrow(new Error('note not found: "missing.md"'))
     })
 
     it("reports a path through a file as not found, never the server's path", async () => {
@@ -790,7 +790,7 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('blockId "nonexistent" not found')
+      ).rejects.toThrow(new Error('blockId "nonexistent" not found in "tasks.md"'))
     })
 
     it("throws when line does not contain a task", async () => {
@@ -808,7 +808,7 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow("no task at line 1")
+      ).rejects.toThrow(new Error("no task at line 1"))
     })
 
     it("throws when no mutations specified", async () => {
@@ -842,7 +842,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('heading "Nonexistent" not found')
+      ).rejects.toThrow(
+        new Error('heading "Nonexistent" not found; available: Active, Up Next, Done'),
+      )
     })
 
     it("throws when multiple done lanes and no explicit heading", async () => {
@@ -988,7 +990,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('hidden path blocked: ".trash/tasks.md" targets a hidden file or folder')
+      ).rejects.toThrow(
+        new Error('hidden path blocked: ".trash/tasks.md" targets a hidden file or folder'),
+      )
       expect(await readTestNote(vault, ".trash/tasks.md")).toBe(SIMPLE_NOTE)
     })
   })
@@ -1199,7 +1203,7 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow('blockId "walk-dog" already exists')
+      ).rejects.toThrow(new Error('blockId "walk-dog" already exists in this note'))
     })
 
     it("errors on invalid block_id characters", async () => {
@@ -1217,7 +1221,11 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow("contains invalid characters")
+      ).rejects.toThrow(
+        new Error(
+          'blockId "bad id!" contains invalid characters (allowed: letters, digits, hyphens)',
+        ),
+      )
     })
 
     it("errors when heading is missing on a Kanban board", async () => {
@@ -1240,7 +1248,26 @@ kanban-plugin: board
       )
     })
 
-    it("errors on invalid date", async () => {
+    it("errors on a malformed date", async () => {
+      const vault = await createVault()
+      await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
+
+      await expect(
+        taskMutations.createTask(
+          {
+            statusRegistry: DEFAULT_STATUS_REGISTRY,
+            vaultPath: vault,
+            path: "tasks.md",
+            description: "Bad date",
+            blockId: "bad-date",
+            due: "2026-2-3",
+          },
+          logger,
+        ),
+      ).rejects.toThrow(new Error('invalid date: due "2026-2-3" (use YYYY-MM-DD)'))
+    })
+
+    it("errors on a date the calendar lacks, saying so rather than naming the format", async () => {
       const vault = await createVault()
       await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
 
@@ -1256,9 +1283,48 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow('invalid date: due "2026-02-30"')
+      ).rejects.toThrow(
+        new Error(
+          'invalid date: due "2026-02-30" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+        ),
+      )
     })
 
+    it.each([
+      {
+        field: "scheduled",
+        dateParams: { scheduled: "2026-02-30" },
+        expectedMessage:
+          'invalid date: scheduled "2026-02-30" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+      },
+      {
+        field: "start",
+        dateParams: { start: "2026-04-31" },
+        expectedMessage:
+          'invalid date: start "2026-04-31" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+      },
+    ])(
+      "errors on a $field date the calendar lacks, naming the field",
+      async ({ dateParams, expectedMessage }) => {
+        const vault = await createVault()
+        await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
+
+        await expect(
+          taskMutations.createTask(
+            {
+              statusRegistry: DEFAULT_STATUS_REGISTRY,
+              vaultPath: vault,
+              path: "tasks.md",
+              description: "Bad date",
+              blockId: "bad-date",
+              ...dateParams,
+            },
+            logger,
+          ),
+        ).rejects.toThrow(new Error(expectedMessage))
+      },
+    )
+
     it("errors when parent not found", async () => {
       const vault = await createVault()
       await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
@@ -1275,7 +1341,7 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow("parent task not found")
+      ).rejects.toThrow(new Error('parent task not found: blockId "nonexistent"'))
     })
 
     it("creates a sub-task under a parent identified by line number", async () => {
@@ -1317,7 +1383,9 @@ kanban-plugin: board
           logger,
         ),
       ).rejects.toThrow(
-        'dependsOn entry " " contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        new Error(
+          'dependsOn entry " " contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        ),
       )
       const content = await readTestNote(vault, "tasks.md")
       expect(content).toBe(SIMPLE_NOTE)
@@ -1340,7 +1408,9 @@ kanban-plugin: board
           logger,
         ),
       ).rejects.toThrow(
-        'taskId "has space" contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        new Error(
+          'taskId "has space" contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        ),
       )
       const content = await readTestNote(vault, "tasks.md")
       expect(content).toBe(SIMPLE_NOTE)
@@ -1509,7 +1579,7 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow("parent task not found: line 1")
+      ).rejects.toThrow(new Error("parent task not found: line 1"))
     })
 
     it("errors on empty description", async () => {
@@ -1938,7 +2008,9 @@ kanban-plugin: board
           logger,
         ),
       ).rejects.toThrow(
-        'dependsOn entry " " contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        new Error(
+          'dependsOn entry " " contains invalid characters (allowed: letters, digits, hyphens, underscores)',
+        ),
       )
       const content = await readTestNote(vault, "tasks.md")
       expect(content).toBe(SIMPLE_NOTE)
@@ -2103,7 +2175,7 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow('blockId "dup" already exists in this note')
+      ).rejects.toThrow(new Error('blockId "dup" already exists in this note'))
       const content = await readTestNote(vault, "tasks.md")
       expect(content).toBe("- [ ] Task A ➕ 2026-01-01 ^dup  \n- [ ] Task B ➕ 2026-01-02 ^other\n")
     })
@@ -2207,7 +2279,7 @@ kanban-plugin: board
       )
     })
 
-    it("errors on invalid date in update", async () => {
+    it("errors on a malformed date in update", async () => {
       const vault = await createVault()
       await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
 
@@ -2222,9 +2294,65 @@ kanban-plugin: board
           },
           logger,
         ),
-      ).rejects.toThrow("invalid date")
+      ).rejects.toThrow(new Error('invalid date: due "not-a-date" (use YYYY-MM-DD)'))
+    })
+
+    it("errors on a date the calendar lacks in update, saying so rather than naming the format", async () => {
+      const vault = await createVault()
+      await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
+
+      await expect(
+        taskMutations.updateTask(
+          {
+            statusRegistry: DEFAULT_STATUS_REGISTRY,
+            vaultPath: vault,
+            path: "tasks.md",
+            blockId: "walk-dog",
+            created: "2026-13-01",
+          },
+          logger,
+        ),
+      ).rejects.toThrow(
+        new Error(
+          'invalid date: created "2026-13-01" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+        ),
+      )
     })
 
+    it.each([
+      {
+        field: "scheduled",
+        dateParams: { scheduled: "2026-02-30" },
+        expectedMessage:
+          'invalid date: scheduled "2026-02-30" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+      },
+      {
+        field: "start",
+        dateParams: { start: "2026-04-31" },
+        expectedMessage:
+          'invalid date: start "2026-04-31" is not a calendar date. Pass a real date in YYYY-MM-DD format.',
+      },
+    ])(
+      "errors on a $field date the calendar lacks in update, naming the field",
+      async ({ dateParams, expectedMessage }) => {
+        const vault = await createVault()
+        await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
+
+        await expect(
+          taskMutations.updateTask(
+            {
+              statusRegistry: DEFAULT_STATUS_REGISTRY,
+              vaultPath: vault,
+              path: "tasks.md",
+              blockId: "walk-dog",
+              ...dateParams,
+            },
+            logger,
+          ),
+        ).rejects.toThrow(new Error(expectedMessage))
+      },
+    )
+
     it("errors on empty description in update", async () => {
       const vault = await createVault()
       await writeTestNote(vault, "tasks.md", SIMPLE_NOTE)
@@ -3320,7 +3448,9 @@ kanban-plugin: board
             logger,
           ),
         ).rejects.toThrow(
-          `cannot reorder within "Tasks" — the heading appears 2 times; rename one section to make it unique`,
+          new Error(
+            `cannot reorder within "Tasks" — the heading appears 2 times; rename one section to make it unique`,
+          ),
         )
       })
 
@@ -3342,7 +3472,9 @@ kanban-plugin: board
             logger,
           ),
         ).rejects.toThrow(
-          `cannot place at position 1 under "Done" — the heading appears 2 times; rename one section to make it unique`,
+          new Error(
+            `cannot place at position 1 under "Done" — the heading appears 2 times; rename one section to make it unique`,
+          ),
         )
       })
 
@@ -3389,7 +3521,9 @@ kanban-plugin: board
             logger,
           ),
         ).rejects.toThrow(
-          `cannot place at position 2 under "Tasks" — the heading appears 2 times; rename one section to make it unique`,
+          new Error(
+            `cannot place at position 2 under "Tasks" — the heading appears 2 times; rename one section to make it unique`,
+          ),
         )
       })
 
@@ -4525,7 +4659,9 @@ title: Tasks
         logger,
       ),
     ).rejects.toThrow(
-      'unrecognized recurrence rule "whenever I remember" (use the Tasks plugin\'s natural language, e.g. "every week", "every 2 weeks when done")',
+      new Error(
+        'unrecognized recurrence rule "whenever I remember" (use the Tasks plugin\'s natural language, e.g. "every week", "every 2 weeks when done")',
+      ),
     )
   })
 
@@ -4673,7 +4809,11 @@ title: Tasks
         },
         logger,
       ),
-    ).rejects.toThrow('unrecognized recurrence rule "sometimes"')
+    ).rejects.toThrow(
+      new Error(
+        'unrecognized recurrence rule "sometimes" (use the Tasks plugin\'s natural language, e.g. "every week", "every 2 weeks when done")',
+      ),
+    )
   })
 
   it("bases the recurrence on the next-priority date when the due date is calendar-invalid", async () => {
@@ -6335,7 +6475,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry')
+      ).rejects.toThrow(
+        new Error('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry'),
+      )
     })
 
     it("updateTask rejects a NON_TASK checkbox by line number", async () => {
@@ -6357,7 +6499,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry')
+      ).rejects.toThrow(
+        new Error('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry'),
+      )
     })
 
     it("createTask rejects a NON_TASK parent by block_id", async () => {
@@ -6380,7 +6524,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry')
+      ).rejects.toThrow(
+        new Error('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry'),
+      )
     })
 
     it("createTask rejects a NON_TASK parent by line number", async () => {
@@ -6403,7 +6549,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry')
+      ).rejects.toThrow(
+        new Error('checkbox "[>]" is a NON_TASK status in the Tasks plugin registry'),
+      )
     })
 
     it("allows updating a normal task when NON_TASK statuses exist", async () => {
@@ -6494,7 +6642,9 @@ title: Tasks
           logger,
         ),
       ).rejects.toThrow(
-        'no checkbox symbol for status "cancelled" in the Tasks plugin registry (the default "-" is typed todo)',
+        new Error(
+          'no checkbox symbol for status "cancelled" in the Tasks plugin registry (the default "-" is typed todo)',
+        ),
       )
     })
   })
@@ -6592,7 +6742,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow('blockId "example" is inside a fenced code block or comment in "tasks.md"')
+      ).rejects.toThrow(
+        new Error('blockId "example" is inside a fenced code block or comment in "tasks.md"'),
+      )
     })
 
     it("createTask rejects a parent inside a fenced code block", async () => {
@@ -6616,7 +6768,9 @@ title: Tasks
           logger,
         ),
       ).rejects.toThrow(
-        'parent task not found: blockId "fenced" is inside a fenced code block or comment',
+        new Error(
+          'parent task not found: blockId "fenced" is inside a fenced code block or comment',
+        ),
       )
     })
 
@@ -6731,7 +6885,7 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow("line 6 is inside a fenced code block or comment")
+      ).rejects.toThrow(new Error("line 6 is inside a fenced code block or comment"))
     })
 
     it("createTask rejects a line-addressed parent inside a fenced code block", async () => {
@@ -6754,7 +6908,9 @@ title: Tasks
           },
           logger,
         ),
-      ).rejects.toThrow("parent task not found: line 6 is inside a fenced code block or comment")
+      ).rejects.toThrow(
+        new Error("parent task not found: line 6 is inside a fenced code block or comment"),
+      )
     })
   })
 })
diff --git a/src/vault-mcp/vault-operations/task-mutations.ts b/src/vault-mcp/vault-operations/task-mutations.ts
index 48ac83602..7157273e7 100644
--- a/src/vault-mcp/vault-operations/task-mutations.ts
+++ b/src/vault-mcp/vault-operations/task-mutations.ts
@@ -6,6 +6,7 @@ import { DateTime } from "luxon"
 import { parseNoteForRewrite, stringifyNote } from "../obsidian-markdown/frontmatter.js"
 import { resolveSafePath, atomicWriteFile } from "./vault-filesystem.js"
 import { assertPathHasExtension } from "../../utils/assert-path-has-extension.js"
+import { classifyIsoDate } from "../../utils/classify-iso-date.js"
 import { readFileOrNull } from "../../utils/fs.js"
 import { withExclusiveFileLock } from "../../utils/file-write-lock.js"
 import { parseHeadings, type HeadingInfo } from "../obsidian-markdown/headings.js"
@@ -1404,11 +1405,20 @@ const assertRecurrenceRuleGrammar = (recurrenceText: string | null | undefined):
  *  reads as part of the task. */
 const TASK_TEXT_LINE_BREAK_PATTERN = /[\r\n]/
 
-/** Validates a date string is a real calendar date. */
-const validateDate = (date: string, fieldName: string): void => {
-  if (!DateTime.fromFormat(date, "yyyy-MM-dd").isValid) {
+/** Rejects a task date param that isn't in YYYY-MM-DD form or names a day the
+ *  calendar lacks. Both messages open with "invalid date", which the
+ *  vault_create_task and vault_update_task Errors sections quote. */
+const assertTaskDate = (date: string, fieldName: string): void => {
+  const dateClass = classifyIsoDate(date)
+
+  if (dateClass === "calendar-date") return
+
+  if (dateClass === "malformed") {
     throw new Error(`invalid date: ${fieldName} "${date}" (use YYYY-MM-DD)`)
   }
+  throw new Error(
+    `invalid date: ${fieldName} "${date}" is not a calendar date. Pass a real date in YYYY-MM-DD format.`,
+  )
 }
 
 type ParentLocator = { kind: "blockId"; blockId: string } | { kind: "line"; line: number }
@@ -1517,15 +1527,14 @@ const createTask = async (params: CreateTaskParams, logger: Logger): Promise<Cre
     throw new Error("description must be a single line")
   }
 
-  // Validate dates
   if (due) {
-    validateDate(due, "due")
+    assertTaskDate(due, "due")
   }
   if (scheduled) {
-    validateDate(scheduled, "scheduled")
+    assertTaskDate(scheduled, "scheduled")
   }
   if (start) {
-    validateDate(start, "start")
+    assertTaskDate(start, "start")
   }
 
   if (parentBlockId && parentLine) {
@@ -1787,7 +1796,6 @@ const updateTask = async (params: UpdateTaskParams, logger: Logger): Promise<Upd
     )
   }
 
-  // Validate dates
   const dateParams: ReadonlyArray<{
     field: DateFieldKey
     value: string | null | undefined
@@ -1797,9 +1805,11 @@ const updateTask = async (params: UpdateTaskParams, logger: Logger): Promise<Upd
     { field: "start", value: start },
     { field: "created", value: created },
   ]
+
   for (const { field, value } of dateParams) {
+    // null clears the date and undefined leaves it as it is; only a string sets one
     if (typeof value === "string") {
-      validateDate(value, field)
+      assertTaskDate(value, field)
     }
   }