Close three validator gaps: savedqueries, AppModuleSiteMap, nested flow actions

CHANGELOG.md

@@ -1,5 +1,23 @@
 # Changelog
 
+## 2026-07-22
+
+### dataverse-xml-lsp 2.2.0
+
+- Validate pac's `<savedqueries>` wrapper root (exported `SavedQueries/*.xml`) by extracting and
+  checking each inner `<savedquery>` against `Fetch.xsd` — indicative, like forms. Previously every
+  exported saved-query file failed as an unknown root.
+- Validate the model-driven-app `<AppModuleSiteMap>` root by extracting its inner `<SiteMap>` and
+  checking it against `SiteMap.xsd` — authoritative.
+- Both new roots are covered by the PostToolUse hook. Removed the stale live
+  `**/SavedQueries/**/*.xml → Fetch.xsd` association, which mis-fired on the `<savedqueries>` root.
+
+### cloud-flow-json-lsp 2.1.0
+
+- The clientdata schema now validates actions nested inside `Scope`/`If`/`Foreach`/`Switch`
+  recursively, so a bogus `runAfter` status or a missing `type` on a nested action is caught instead
+  of passing silently. Top-level behaviour is unchanged.
+
 ## 2.1.0 — 2026-07-21
 
 ### dataverse-xml-lsp

README.md

@@ -1,33 +1,86 @@
 # dataverse-agent-plugins
 
-Agent plugins for Microsoft Dataverse / Power Platform development. A
-[Claude Code plugin marketplace](https://docs.anthropic.com/en/docs/claude-code) — each
-plugin also works standalone (plain PowerShell scripts) for other agents, VS Code, and CI.
+Catch broken Dataverse / Power Platform customization files **while you edit them** — before
+`pac solution import` fails at the end of a long round-trip.
+
+This is a [Claude Code plugin marketplace](https://docs.anthropic.com/en/docs/claude-code) with
+two language-server (LSP) plugins. Each one validates a kind of hand-edited solution file against a
+real schema and shows the errors inline as you type. Every plugin also runs **standalone** — plain
+PowerShell/Node with no agent — so the same checks work from other agents, VS Code, or CI.
+
+## Plugins
+
+| Plugin | Validates | How |
+|--------|-----------|-----|
+| [dataverse-xml-lsp](plugins/dataverse-xml-lsp/) | Dataverse customization XML — ribbon (`RibbonDiff.xml`), sitemap, app sitemap, forms, saved queries, FetchXML, charts, ISV config, `Customizations.xml` | Live [lemminx](https://github.com/eclipse/lemminx) LSP diagnostics against the official Microsoft XSDs, plus a standalone PowerShell validator (`Validate-DataverseXml.ps1`) for wrapper files and CI |
+| [cloud-flow-json-lsp](plugins/cloud-flow-json-lsp/) | Unpacked Power Automate cloud-flow clientdata (`Workflows/*.json`) | Live `vscode-json-language-server` diagnostics against a bundled clientdata/WDL wrapper schema; headless checks via built-in `Test-Json` |
+
+## When would I use this?
+
+- **You edit solution files as code.** You unpack a solution (`pac solution unpack`), tweak a ribbon
+  button, a form, a sitemap, or a cloud flow by hand, then pack and import. These plugins turn the
+  errors that would otherwise only appear at import time into red squiggles while you edit.
+- **An agent is editing them for you.** In Claude Code (or Codex), the validator runs after edits so
+  the agent sees its own mistakes and fixes them, instead of confidently producing a file that won't
+  import.
+- **You want a CI gate.** Run the standalone validator over changed files in a pipeline so a bad edit
+  fails the build, not the deployment.
+
+Neither plugin talks to Dataverse or needs auth — they validate files on disk. A successful
+`pac solution import` / `pac solution check` remains the final authority; these catch the large
+class of mistakes you don't need a live environment to find.
 
 ## Install (Claude Code)
 
+Add the marketplace once:
+
     /plugin marketplace add Cordedmink2/dataverse-agent-plugins
+
+Then install whichever plugin you need and run its one-time, per-machine setup (this fetches the
+Microsoft XSDs / lemminx / JSON language server — nothing bulky is committed to the repo):
+
     /plugin install dataverse-xml-lsp@dataverse-agent-plugins
     /dataverse-xml-lsp:dataverse-xml-lsp-setup
 
-Update later with `/plugin marketplace update dataverse-agent-plugins` then
-`/plugin update dataverse-xml-lsp@dataverse-agent-plugins` (re-run setup after updates).
+    /plugin install cloud-flow-json-lsp@dataverse-agent-plugins
+    /cloud-flow-json-lsp:cloud-flow-json-lsp-setup
 
-## Plugins
+Run `/reload-plugins` afterwards. Update later with
+`/plugin marketplace update dataverse-agent-plugins`, then
+`/plugin update <plugin>@dataverse-agent-plugins`, and re-run that plugin's setup.
+
+## Use it without Claude Code
 
-| Plugin | What it does |
-|--------|--------------|
-| [dataverse-xml-lsp](plugins/dataverse-xml-lsp/) | LSP + CLI validator for Dataverse customization XML (ribbon, sitemap, forms, FetchXML, charts, ISV config and more) against the official Microsoft XSDs. Live lemminx LSP diagnostics + a standalone PowerShell validator. Also usable from [Codex](plugins/dataverse-xml-lsp/docs/codex.md), [VS Code with no agent](plugins/dataverse-xml-lsp/docs/vscode.md), and CI. |
-| [cloud-flow-json-lsp](plugins/cloud-flow-json-lsp/) | JSON LSP for unpacked Power Automate cloud-flow clientdata (`Workflows/*.json`), validated against a bundled clientdata/WDL wrapper schema. Live `vscode-json-language-server` LSP diagnostics; headless structure checks via built-in `Test-Json`. Also usable from [Codex](plugins/cloud-flow-json-lsp/docs/codex.md), [VS Code with no agent](plugins/cloud-flow-json-lsp/docs/vscode.md), and CI. |
+Each plugin works from other tools — see its docs:
+
+- **Other agents (Codex, etc.):** `plugins/<plugin>/docs/codex.md`
+- **VS Code, no agent:** `plugins/<plugin>/docs/vscode.md`
+- **CI / headless:** the standalone validator (`Validate-DataverseXml.ps1`) or `Test-Json`; see the
+  same docs.
 
 ## Requirements
 
-PowerShell 7+ (`pwsh`) on Windows, macOS or Linux. The `cloud-flow-json-lsp` plugin also
-needs Node.js (for its `npm`-installed JSON language server). Plugins fetch what else they need
-at setup (Microsoft XSDs, lemminx binary, the JSON language server) — nothing bulky or
-third-party is committed here.
+PowerShell 7+ (`pwsh`) on Windows, macOS, or Linux. `cloud-flow-json-lsp` also needs Node.js (for its
+`npm`-installed JSON language server). Setup fetches the rest.
+
+## Roadmap
+
+Planned, not yet built (design detail in
+[the validator-gaps spec](docs/superpowers/specs/2026-07-22-validator-gaps-design.md)):
+
+- **Authoritative saved-query validation.** Saved queries currently validate *indicatively* — the
+  bundled Microsoft `Fetch.xsd` lags real exports (`layoutxml`, `LocalizedNames`), so a few
+  expected "not declared" errors are noise, the same as whole-form `FormXml`. A setup-time patch
+  step would extend the schema so saved queries validate cleanly and the patch survives an XSD
+  re-download.
+- **Semantic cloud-flow linting.** A layer above the JSON shape check that catches things a schema
+  can't express: `runAfter` naming a non-sibling action or a cycle, a `connectionName` that resolves
+  to no declared connection reference (the top cause of a flow importing turned Off), child-invoker
+  connection mistakes, hard-coded environment GUIDs / host / site URLs, and condition rows that are
+  empty or always-true (`{"equals":["",""]}`). Bundled into `cloud-flow-json-lsp` as its "semantics"
+  layer.
 
 ## License
 
-MIT. The Microsoft XSD schemas are downloaded from Microsoft at setup time and remain
-subject to Microsoft's terms — they are not redistributed in this repo.
+MIT. The Microsoft XSD schemas are downloaded from Microsoft at setup time and remain subject to
+Microsoft's terms — they are not redistributed in this repo.

docs/superpowers/specs/2026-07-22-validator-gaps-design.md

@@ -0,0 +1,148 @@
+# Close three validator gaps (dataverse-xml-lsp + cloud-flow-json-lsp)
+
+Date: 2026-07-22
+Status: Approved for planning
+
+## Background
+
+Testing both plugins against a real client solution export (712 customization XML files
+and 12 unpacked cloud flows) surfaced three gaps where a validator silently fails to
+cover the most common real-export shapes. Each finding below was reproduced against the
+installed plugins (cache paths, freshly downloaded schemas, LSP servers reloaded), so the
+gaps are in the mapping/schema logic, not in setup.
+
+| Finding | Plugin | Real-data evidence |
+|---------|--------|--------------------|
+| `<savedqueries>` wrapper root rejected | dataverse-xml-lsp | 295 / 712 files fail root detection |
+| `<AppModuleSiteMap>` root rejected | dataverse-xml-lsp | modern per-app sitemaps uncovered |
+| Nested-action errors not caught | cloud-flow-json-lsp | bogus `runAfter` in a nested action passes silently |
+
+Two design facts established during investigation drive the approach:
+
+- `AppModuleSiteMap` contains an inner `<SiteMap>` element carrying the Area/Group/SubArea
+  tree. Extracting and validating that inner element against `SiteMap.xsd` produces **0
+  errors** on real data — so this fix is authoritative, not indicative.
+- A real `<savedquery>` fragment carries `layoutxml` / `LocalizedNames` (absent from the
+  bundled `Fetch.xsd`) and omits elements the schema marks required. Validation is
+  therefore **indicative** — the same situation as FormXml today.
+
+## Goals
+
+- Real exported saved queries and per-app sitemaps validate instead of being rejected at
+  root detection.
+- Cloud-flow structural validation reaches actions nested inside Scope / If / Foreach /
+  Switch, not just top-level actions.
+- Every fix ships with a regression test derived from the real-export shapes.
+
+## Non-goals
+
+- Making saved-query validation authoritative (patching the Microsoft XSD) — deferred to
+  the roadmap.
+- Semantic flow linting (condition rows, connection-reference resolution, etc.) — deferred
+  to the roadmap.
+- The minor "directory input is shallow / flags Entity.xml" polish — out of scope.
+
+## Design
+
+All three changes are small; two reuse existing machinery.
+
+### Fix A — `dataverse-xml-lsp`: `<savedqueries>` wrapper root
+
+Root cause: `Validate-DataverseXml.ps1` maps the singular inner element `savedquery`, but
+pac exports the file with a `<savedqueries>` wrapper, so root detection reports
+`Unknown root element <savedqueries>`.
+
+Changes:
+
+- `scripts/Validate-DataverseXml.ps1`: add `innerElementByRoot['savedqueries'] =
+  'savedquery'`. This reuses the existing per-fragment extraction loop (identical code path
+  to `forms` -> `systemform/form`); no new logic. Each `<savedquery>` child is validated
+  against `Fetch.xsd`.
+- `hooks/validate-wrapper.mjs`: add `savedqueries` to `OWNED_ROOTS` so the PostToolUse hook
+  runs the CLI validator after an edit.
+- `scripts/lsp-launch.mjs`: remove the stale live association `**/SavedQueries/**/*.xml ->
+  Fetch.xsd`. It mis-fires: lemminx validates the whole document, whose root
+  `<savedqueries>` is not declared in `Fetch.xsd`, producing a misleading root-level error.
+  Coverage moves to the hook, consistent with the other wrapper/lag-prone roots.
+
+Behaviour: **indicative**. The known OOB errors (`layoutxml` not declared; required
+`name` / `returnedtypecode` reported missing) are expected noise, read the same way as
+FormXml — confirm your own edit is not named in the output.
+
+### Fix B — `dataverse-xml-lsp`: `<AppModuleSiteMap>` wrapper root
+
+Root cause: model-driven-app sitemaps export with root `<AppModuleSiteMap>`, which is not
+in the root map; the `**/SiteMap*.xml` glob does not match the filename either.
+
+Changes:
+
+- `scripts/Validate-DataverseXml.ps1`: add `rootToSchema['AppModuleSiteMap'] =
+  'SiteMap.xsd'` and `innerElementByRoot['AppModuleSiteMap'] = 'SiteMap'`. The inner
+  `<SiteMap>` subtree is extracted and validated.
+- `hooks/validate-wrapper.mjs`: add `AppModuleSiteMap` to `OWNED_ROOTS`.
+
+Behaviour: **authoritative** (0 errors on real data). No live LSP association — lemminx
+cannot do per-fragment extraction, so this follows the wrapper-root-via-hook pattern.
+
+### Fix C — `cloud-flow-json-lsp`: recursive nested-action validation
+
+Root cause: `schemas/cloud-flow-clientdata.schema.json` constrains only the top-level
+`actions` map. Actions nested inside a container (`Scope` / `If` / `Foreach` / `Switch`)
+fall under `additionalProperties: true` and are unchecked, so a bogus nested `runAfter`
+status or a missing nested `type` passes silently.
+
+Changes:
+
+- Hoist the action object shape into `definitions/action` (draft-07 keyword `definitions`,
+  not `$defs`, so the bundled and VS Code validators both resolve it).
+- `$ref` `#/definitions/action` from `definition.actions.additionalProperties` and,
+  recursively, from the container keys inside an action: `actions`, `else.actions`,
+  `cases.*.actions`, `default.actions`.
+- Keep `inputs` untyped (the intentional looseness that avoids OpenApiConnection false
+  positives). Only `type` presence and the `runAfter` status enum are enforced, now at any
+  depth.
+
+## Testing
+
+Each plugin gains fixtures derived from the real-export shapes, kept minimal:
+
+- dataverse-xml-lsp: a valid `<savedqueries>` file (passes with only the documented OOB
+  noise), a valid `<AppModuleSiteMap>` file (passes clean), and an invalid variant of each.
+- cloud-flow-json-lsp: a valid nested-action flow and an invalid one with a bogus
+  `runAfter` status on a nested action — which must now be caught.
+
+Assertions extend the existing suites:
+
+- `dataverse-xml-lsp/tests/Validate-DataverseXml.Tests.ps1`: new roots resolve to the right
+  schema and inner element; valid/invalid fixtures behave as expected.
+- `dataverse-xml-lsp/tests/Hook.Tests.ps1`: the hook now claims `savedqueries` and
+  `AppModuleSiteMap`.
+- `dataverse-xml-lsp` LSP association test: the `SavedQueries` association is gone.
+- `cloud-flow-json-lsp`: the setup self-check / smoke test gains the nested-bad-runAfter
+  case and confirms a diagnostic fires.
+
+Committed fixtures stay minimal and synthetic (derived from the real shapes, not client
+data). Two real client exports — `Claude/QEII` and `Claude/NZLS` — are available locally as
+additional corpora for a broad pre-release sweep of each fix; they are not committed.
+
+## Docs and versioning
+
+- `dataverse-xml-lsp/docs/guide.md`: add both new roots to the root->schema table; extend
+  the "indicative, not authoritative" note to name `savedqueries`.
+- `CHANGELOG.md`: one entry per plugin.
+- Version bumps: `dataverse-xml-lsp` 2.1.0 -> 2.2.0; `cloud-flow-json-lsp` 2.0.0 -> 2.1.0.
+
+## Roadmap (documented, not built here)
+
+1. **Authoritative saved queries.** Add a post-download patch step in
+   `dataverse-xml-lsp/scripts/Get-Schemas.ps1` that extends the stock `Fetch.xsd` (declare
+   `layoutxml` / `LocalizedNames`, relax the required attributes) so saved queries validate
+   cleanly rather than indicatively, and the patch survives XSD re-download.
+2. **Flow semantic-validate skill.** Port the checks from the personal
+   `power-automate-flow-dev` skill (`flow-lint.ps1` + `flow-validate-conditions.ps1`) into
+   `cloud-flow-json-lsp` as a bundled semantic layer above the shape schema: `runAfter`
+   integrity and cycles, `connectionName` -> `connectionReferences` resolution,
+   child-invoker connection detection, hardcoded env GUID/host/site-URL warnings, and
+   condition-row checks (empty and/or groups, empty operands, the always-true
+   `{"equals":["",""]}` row, stringified literals). This is the "Semantics" layer the guide
+   already references as living in an external skill.

llms.txt

@@ -13,7 +13,9 @@
 
 - [Repo README](README.md): marketplace overview, install, requirements
 - [Plugin README](plugins/dataverse-xml-lsp/README.md): full setup for every consumer type
-- [SKILL.md](plugins/dataverse-xml-lsp/SKILL.md): the agent skill — validation workflow, root-element→schema table, ribbon-button recipe, gotchas
+- [validate skill](plugins/dataverse-xml-lsp/skills/dataverse-xml-validate/SKILL.md): when and how to run the validator
+- [setup skill](plugins/dataverse-xml-lsp/skills/dataverse-xml-lsp-setup/SKILL.md): one-time per-machine install
+- [guide.md](plugins/dataverse-xml-lsp/docs/guide.md): root-element→schema table (incl. savedqueries and AppModuleSiteMap wrapper roots), ribbon-button recipe, gotchas
 - [Codex guide](plugins/dataverse-xml-lsp/docs/codex.md): using the validator loop from non-Claude agents
 - [VS Code guide](plugins/dataverse-xml-lsp/docs/vscode.md): editor-only setup, CI usage
 - [Debugging](plugins/dataverse-xml-lsp/docs/debugging.md): setup failures, missing diagnostics, false positives
@@ -22,13 +24,15 @@
 ### cloud-flow-json-lsp
 
 - [Plugin README](plugins/cloud-flow-json-lsp/README.md): install + setup for every consumer type
-- [SKILL.md](plugins/cloud-flow-json-lsp/SKILL.md): the shape-vs-semantics split, what attaches, gotchas
+- [SKILL.md](plugins/cloud-flow-json-lsp/SKILL.md): one-time per-machine install
+- [guide.md](plugins/cloud-flow-json-lsp/docs/guide.md): the shape-vs-semantics split, what attaches, recursive nested-action validation, gotchas
 - [Codex guide](plugins/cloud-flow-json-lsp/docs/codex.md): the `Test-Json` validation loop from non-Claude agents
 - [VS Code guide](plugins/cloud-flow-json-lsp/docs/vscode.md): editor-only setup, CI usage
 - [Debugging](plugins/cloud-flow-json-lsp/docs/debugging.md): setup failures, missing diagnostics
 - [Schema source](plugins/cloud-flow-json-lsp/schemas/SOURCE.md): what the wrapper schema validates and why
 
 ## Optional
 
-- [Design spec](docs/superpowers/specs/2026-07-15-dataverse-agent-plugins-repo-design.md): why the repo is shaped this way
+- [Repo design spec](docs/superpowers/specs/2026-07-15-dataverse-agent-plugins-repo-design.md): why the repo is shaped this way
+- [Validator-gaps spec](docs/superpowers/specs/2026-07-22-validator-gaps-design.md): savedqueries + AppModuleSiteMap + nested-flow fixes and the roadmap behind them
 - [Changelog](CHANGELOG.md)

plugins/cloud-flow-json-lsp/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://anthropic.com/claude-code/plugin.schema.json",
   "name": "cloud-flow-json-lsp",
-  "version": "2.0.0",
+  "version": "2.1.0",
   "description": "JSON LSP for unpacked Power Automate cloud-flow clientdata: live vscode-json-language-server diagnostics on solution Workflows/*.json against a bundled clientdata/WDL wrapper schema.",
   "author": {
     "name": "Connor Parsons"

plugins/cloud-flow-json-lsp/docs/guide.md

@@ -13,7 +13,7 @@ It is one layer of a two-layer story — do not confuse them:
 
 | Layer | What it checks | Where |
 |-------|----------------|-------|
-| **Shape (this plugin)** | JSON well-formedness + the clientdata/WDL wrapper structure: `properties.definition` present, `definition` has `$schema`/`triggers`/`actions`, `runAfter` statuses are the WDL enum. Live, in-editor. | `schemas/cloud-flow-clientdata.schema.json` via `vscode-json-language-server` (`.lsp.json`) |
+| **Shape (this plugin)** | JSON well-formedness + the clientdata/WDL wrapper structure: `properties.definition` present, `definition` has `$schema`/`triggers`/`actions`, `runAfter` statuses are the WDL enum, validated recursively through Scope/If/Foreach/Switch. Live, in-editor. | `schemas/cloud-flow-clientdata.schema.json` via `vscode-json-language-server` (`.lsp.json`) |
 | **Semantics (the `power-automate-flow-dev` skill)** | Cross-node rules JSON Schema *cannot* express: `runAfter` naming a real sibling, `connectionName` resolving to a declared connection reference, child-invoker connections, hard-coded environment values, condition rows. Plus the export → unpack → edit → pack → import → verify round-trip. | that skill's `flow-lint.ps1` |
 
 **Rule of thumb:** the LSP catches "this isn't a well-formed flow file." `flow-lint.ps1` catches "this