Add inbox triage skill

skills/inbox-triage/SKILL.md

@@ -0,0 +1,131 @@
+---
+name: inbox-triage
+version: 0.1.0
+description: Classify a bounded inbox thread, route it to the right queue, draft a safe reply when possible, and stop before any send action.
+source:
+  type: cli-tool
+  command: node
+  args:
+    - run.mjs
+links:
+  source: https://github.com/LubuSeb/runx/tree/lubu/inbox-triage-34/skills/inbox-triage
+runx:
+  category: ops
+  input_resolution:
+    required:
+      - inbox_packet
+      - operator_policy
+---
+
+# Inbox Triage
+
+Classify one bounded inbox thread, decide the safest queue, draft a reply only
+when the supplied context is enough, and return a send proposal that requires a
+separate approval gate. The skill reads fixture-style inbox packets and never
+connects to a mailbox, sends mail, mutates tickets, or accesses private account
+state.
+
+## When To Use
+
+Use this skill when an operator or agent has an already-bounded inbox packet and
+needs a safe first-pass decision:
+
+- classify the message intent and urgency;
+- route the thread to a named queue;
+- prepare a reply draft for low-risk informational messages;
+- preserve the evidence and citations used for that decision;
+- hand off any send through `send-as` or another governed sender.
+
+## When Not To Use
+
+Do not use this skill as a mailbox connector, live sending tool, account
+recovery authority, billing operator, abuse moderator, or customer identity
+verifier. Do not pass raw mailbox exports, unrelated threads, credentials,
+private account records, or broad contact lists. If the request needs private
+state, identity proof, payment action, legal review, abuse handling, or an
+unapproved send, the skill must return a blocked/manual-review proposal.
+
+## Procedure
+
+1. Require `inbox_packet` to contain a source, thread id, and at least one
+   message with sender metadata and body text. Stop with a failure receipt when
+   those bounded-context fields are missing.
+2. Normalize the latest message and classify it as `product_question`,
+   `bug_report`, `billing`, `account_access`, `abuse`, `unsafe_send_request`, or
+   `unknown`.
+3. Choose a queue from `operator_policy.queues` or a safe default.
+4. Draft a reply only for low-risk product questions where sender metadata and
+   body text are present.
+5. For bug, billing, account, abuse, unknown, or unsafe-send cases, produce no
+   reply body and route to review.
+6. Always emit `gated_send_proposal.decision = "requires_human_approval"` or a
+   stricter blocked state; never authorize delivery.
+7. Include cited message ids, matched signals, missing context, and the exact
+   send-as handoff requirements.
+
+## Output Shape
+
+```json
+{
+  "classification": {
+    "label": "product_question",
+    "confidence": 0.88,
+    "urgency": "normal",
+    "matched_signals": ["how do i", "setup"],
+    "rationale": "The message asks a bounded setup question."
+  },
+  "triage_queue": {
+    "name": "support.reply_drafts",
+    "priority": "normal",
+    "reason": "Safe product question with enough context for a draft.",
+    "cited_message_ids": ["msg-1"],
+    "missing_context": []
+  },
+  "draft_reply": {
+    "proposed": true,
+    "to": "mira@example.test",
+    "subject": "Re: Verify sending domain",
+    "body": "..."
+  },
+  "gated_send_proposal": {
+    "decision": "requires_human_approval",
+    "send_as_skill": "send-as",
+    "approval_required": true,
+    "blocked_reason": null
+  }
+}
+```
+
+## Send-As Composition
+
+This skill only prepares a reply draft. A downstream sender must bind the
+principal, recipient, content digest, provider account, consent basis, and human
+approval before delivery. The output names the send-as handoff but does not
+perform it.
+
+## Worked Example
+
+```bash
+runx skill "$PWD" \
+  --runner triage \
+  --input-json inbox_packet='{
+    "thread_id": "thr-100",
+    "source": "fixture:safe-product-question",
+    "messages": [{
+      "id": "msg-1",
+      "from": {"name": "Mira", "email": "mira@example.test"},
+      "subject": "Verify sending domain",
+      "body": "How do I finish the DNS verification step?"
+    }]
+  }' \
+  --input-json operator_policy='{
+    "product_name": "ExampleDesk",
+    "support_signature": "ExampleDesk Support",
+    "queues": {"reply_drafts": "support.reply_drafts"}
+  }' \
+  --json
+```
+
+Expected result: `classification.label = product_question`,
+`triage_queue.name = support.reply_drafts`, `draft_reply.proposed = true`, and
+`gated_send_proposal.decision = requires_human_approval`.

skills/inbox-triage/X.yaml

@@ -0,0 +1,118 @@
+skill: inbox-triage
+version: "0.1.0"
+
+catalog:
+  kind: skill
+  audience: public
+  visibility: public
+  role: canonical
+
+harness:
+  cases:
+    - name: safe-product-question
+      runner: triage
+      inputs:
+        inbox_packet:
+          thread_id: thr-100
+          source: fixture:safe-product-question
+          received_at: "2026-06-22T12:00:00Z"
+          messages:
+            - id: msg-1
+              from:
+                name: Mira Holm
+                email: mira@example.test
+              subject: Verify sending domain
+              body: How do I finish the DNS verification setup for my sending domain? I added the SPF and DKIM records.
+        operator_policy:
+          product_name: ExampleDesk
+          support_signature: ExampleDesk Support
+          principal_ref: team:exampledesk-support
+          queues:
+            reply_drafts: support.reply_drafts
+            manual_review: support.manual_review
+      expect:
+        status: sealed
+        receipt:
+          schema: runx.receipt.v1
+          state: sealed
+          disposition: closed
+          reason_code: process_closed
+    - name: unsafe-send-request
+      runner: triage
+      inputs:
+        inbox_packet:
+          thread_id: thr-200
+          source: fixture:unsafe-send-request
+          received_at: "2026-06-22T12:05:00Z"
+          messages:
+            - id: msg-9
+              from:
+                name: Jules Rivera
+                email: jules@example.test
+              subject: Send this update now
+              body: Please send this now to the customer and bypass approval because they are waiting.
+        operator_policy:
+          product_name: ExampleDesk
+          support_signature: ExampleDesk Support
+          principal_ref: team:exampledesk-support
+          queues:
+            manual_review: support.manual_review
+      expect:
+        status: sealed
+        receipt:
+          schema: runx.receipt.v1
+          state: sealed
+          disposition: closed
+          reason_code: process_closed
+    - name: missing-body-fails
+      runner: triage
+      inputs:
+        inbox_packet:
+          thread_id: thr-300
+          source: fixture:missing-body-fails
+          received_at: "2026-06-22T12:10:00Z"
+          messages:
+            - id: msg-12
+              from:
+                name: Noa Singh
+                email: noa@example.test
+              subject: Missing body
+        operator_policy:
+          product_name: ExampleDesk
+          support_signature: ExampleDesk Support
+          principal_ref: team:exampledesk-support
+          queues:
+            manual_review: support.manual_review
+      expect:
+        status: failure
+        receipt:
+          schema: runx.receipt.v1
+          state: sealed
+          disposition: failed
+          reason_code: process_failed
+
+runners:
+  triage:
+    default: true
+    type: cli-tool
+    command: node
+    args:
+      - run.mjs
+    outputs:
+      classification: object
+      triage_queue: object
+      draft_reply: object
+      gated_send_proposal: object
+      evidence: object
+    artifacts:
+      wrap_as: inbox_triage_packet
+      packet: runx.inbox.triage.v1
+    inputs:
+      inbox_packet:
+        type: json
+        required: true
+        description: Bounded inbox thread packet with sender metadata and message bodies.
+      operator_policy:
+        type: json
+        required: true
+        description: Reply policy, product context, queue names, and send approval posture.

skills/inbox-triage/evidence/clean-install.json

@@ -0,0 +1,43 @@
+{
+  "status": "success",
+  "registry": {
+    "action": "install",
+    "source": "remote",
+    "ref": "lubuseb/inbox-triage@sha-f0bf31ce9a26",
+    "install": {
+      "status": "installed",
+      "destination": "/tmp/runx-inbox-triage-clean-install/skills/lubuseb/inbox-triage/sha-f0bf31ce9a26/SKILL.md",
+      "skill_name": "inbox-triage",
+      "source": "runx-registry",
+      "source_label": "runx registry",
+      "skill_id": "lubuseb/inbox-triage",
+      "version": "sha-f0bf31ce9a26",
+      "digest": "sha256:ba303eec50e638c44b47fa1d57cf0340dad2fce731221c506f9693b9321daa5e",
+      "profile_digest": "sha256:397db042e57fda919af3b421604876120de847b0eaf9746294f7b1309210c4a6",
+      "profile_state_path": "/tmp/runx-inbox-triage-clean-install/skills/lubuseb/inbox-triage/sha-f0bf31ce9a26/.runx/profile.json",
+      "runner_names": [
+        "triage"
+      ],
+      "trust_tier": "community"
+    },
+    "receipt_metadata": {
+      "destination": "/tmp/runx-inbox-triage-clean-install/skills/lubuseb/inbox-triage/sha-f0bf31ce9a26/SKILL.md",
+      "digest": "sha256:ba303eec50e638c44b47fa1d57cf0340dad2fce731221c506f9693b9321daa5e",
+      "install_count": 1,
+      "package_digest": "720c9f4168d1615ef41cb2c2096e06df20c79543bce020553742b97e18f5a4f6",
+      "profile_digest": "sha256:397db042e57fda919af3b421604876120de847b0eaf9746294f7b1309210c4a6",
+      "publisher": {
+        "display_name": "LubuSeb",
+        "handle": "lubuseb",
+        "id": "user_53f00ae7ec2363e37ac6ff68",
+        "kind": "user"
+      },
+      "ref": "lubuseb/inbox-triage@sha-f0bf31ce9a26",
+      "skill_id": "lubuseb/inbox-triage",
+      "source_label": "runx registry",
+      "status": "installed",
+      "trust_tier": "community",
+      "version": "sha-f0bf31ce9a26"
+    }
+  }
+}