@lifeaitools/rdc-skills 0.9.28 → 0.9.29

package/.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "rdc",
-  "version": "0.9.25",
+  "version": "0.9.28",
   "description": "RDC typed-agent dispatch skill suite for Claude Code — plan, build, review, overnight unattended builds with work-item tracking and TDD enforcement.",
   "author": {
     "name": "LIFEAI",

package/guides/agents/verify.md

@@ -37,6 +37,23 @@ If you're about to write any of these, STOP:
 
 Run these in order. All must pass. Capture output.
 
+### 0. Checklist decomposition quality
+
+Before code verification, inspect each work item checklist and the matching plan section.
+
+Required plan evidence:
+- `## Checklist Decomposition Matrix`
+- `## Checklist Quality Gate`
+- `verdict: PASS`
+
+Required work item evidence:
+- At least one `decomp-*` checklist item for every implementation work item
+- At least one `test-*` checklist item for every implementation work item
+- Each `decomp-*` item includes a route or file path, one action, one expected result, and one evidence artifact
+- Each `decomp-*` item is independently pass/fail; vague rows like "UI complete", "theme management works", or "integration verified" are failures
+
+If this check fails, stop validation for that item and reopen it. Do not mark a work item `done` when the checklist itself is not decomposed enough to prove completion.
+
 ### 1. Per-package vitest (every package touched)
 
 ```bash
@@ -107,6 +124,7 @@ $ npx tsc --noEmit --project packages/hail/tsconfig.json
 (no output — clean)
 
 ### Status
+Checklist quality: PASS
 PASS — safe to mark tasks done.
 ```
 
@@ -124,6 +142,7 @@ If any step fails: do NOT claim completion. Report the failure, fix it, re-run t
 When dispatching a verification agent, give them:
 - Exact list of packages/apps to verify
 - Explicit ban on `pnpm build` / `pnpm test` / `pnpm -r`
+- Work item checklists and the matching plan doc so `decomp-*` quality can be verified first
 - Required output format (above)
 - Instruction: "If any command fails, STOP and report. Do not fix. Do not continue."
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lifeaitools/rdc-skills",
-  "version": "0.9.28",
+  "version": "0.9.29",
   "description": "RDC typed-agent dispatch skill suite for Claude Code — plan, build, review, overnight builds",
   "keywords": [
     "claude-code",

package/skills/build/SKILL.md

@@ -97,6 +97,10 @@ Read the task title and description, then:
    |-----------|--------|
    | No child tasks returned | → Invoke `rdc:plan` on this epic. Do NOT proceed with build. |
    | Tasks exist but all have empty `description` fields | → Invoke `rdc:plan` on this epic. Tasks without descriptions cannot be safely dispatched. |
+   | Plan doc missing `## Checklist Decomposition Matrix` | → Invoke `rdc:plan` on this epic. Do NOT dispatch agents. |
+   | Plan doc missing `## Checklist Quality Gate` with `verdict: PASS` | → Invoke `rdc:plan` on this epic. Do NOT dispatch agents. |
+   | Any implementation task lacks `decomp-*` checklist items | → Invoke `rdc:plan` on this epic. Coarse checklists cannot be safely dispatched. |
+   | Any `decomp-*` item lacks route/file, action, expected result, or evidence artifact | → Invoke `rdc:plan` on this epic. Do NOT dispatch agents. |
    | Tasks exist and have descriptions | → Continue with build. |
 
    **Re-planning is not a failure — it is correct behavior.** The build skill is the last gate before agent dispatch; catching an under-specified epic here is cheaper than a wasted agent run.
@@ -126,17 +130,53 @@ Read the task title and description, then:
    
    **If design decisions exist: follow them.** Include the summary in the agent prompt.
 
-3. **Load the plan** (if exists): check `.rdc/plans/` for matching topic (fallback: `.rdc/plans/`).
+3. **Load the plan** (mandatory): check `.rdc/plans/` for matching topic (fallback: `.rdc/plans/`).
+
+3b. **Checklist decomposition quality gate (mandatory before code):**
+
+   `rdc:build` is the final gate before implementation. It MUST reject any plan or work item that is too coarse for an agent to execute and verify independently.
+
+   Required plan sections:
+   - `## Checklist Decomposition Matrix`
+   - `## Checklist Quality Gate`
+   - `verdict: PASS`
+
+   Required task checklist shape:
+   - Every implementation work item has at least one `decomp-*` checklist row and one `test-*` checklist row.
+   - Every `decomp-*` row names a concrete route or file path.
+   - Every `decomp-*` row names one user/agent action.
+   - Every `decomp-*` row names one expected UI/API/DB result.
+   - Every `decomp-*` row names one verification artifact: test name, route probe, Playwright screenshot, SQL query, API response, type-check, migration proof, or CLI transcript.
+
+   Atomicity rubric:
+   - One observable behavior per `decomp-*` row.
+   - The row can pass or fail without reading hidden intent from the plan narrative.
+   - The row is small enough for a worker to implement, tick, and cite evidence for directly.
+   - Vague rows such as "theme management works", "UI implemented", "verified", "all screens", or "integration complete" are failures.
+
+   Minimum decomposition heuristics:
+   - UI route: at least empty/loading/loaded/error or the documented reason a state does not apply.
+   - CRUD surface: at least list, create, edit, detail, duplicate/delete/archive where applicable, and validation failure.
+   - API route: at least successful read/write, validation failure, and unauthorized/forbidden or documented auth bypass.
+   - DB/migration task: at least schema object, relationship/guard, policy/permission, seed/fixture or backfill, and smoke query.
+   - Editor/sidebar/CLI workflow: at least start, attach/open, enqueue action, observe result, timeout/error, and live refresh where applicable.
+
+   If the gate fails:
+   - Interactive: show the failing rows and invoke `rdc:plan <epic-id>` to repair the matrix before dispatch.
+   - Unattended: invoke `rdc:plan <epic-id> --unattended`, reload tasks, and run this gate once more.
+   - If the second gate fails, abort the build and return the failure list. Do not dispatch implementation agents.
+
+   **Plan verifier escalation:** A separate verifier agent is optional, not default. Dispatch one only when the plan touches 5+ UI routes, 3+ data/API boundaries, auth/security, production deployment, or when this rubric fails twice. The verifier reads the plan and work-item checklists only; it does not write code.
 
 4. **Read CLAUDE.md files** for all affected packages.
 
 5. **Classify each task** → assign agent type from the table above.
 
-5b. **Write a checklist into every work item before dispatching:**
-    For each task, append a checklist to its notes BEFORE setting to `in_progress`:
+5b. **Write or refresh the checklist into every work item before dispatching:**
+    For each task, append the exact `decomp-*` and `test-*` checklist rows to its notes BEFORE setting to `in_progress`:
     ```sql
     SELECT update_work_item_status('<id>'::uuid, 'in_progress',
-      '["CHECKLIST: [ ] <deliverable 1>, [ ] <deliverable 2>, [ ] tsc clean, [ ] route 200, [ ] committed"]'::jsonb
+      '["CHECKLIST: [ ] decomp-ui-route-state: <route/file> | action=<action> | expect=<result> | evidence=<artifact>, [ ] test-smoke-route: <route probe>, [ ] committed"]'::jsonb
     );
     ```
     The agent must complete every item on this checklist and return it checked off in AGENT_COMPLETE.
@@ -191,6 +231,13 @@ Read the task title and description, then:
      - **`"When done, set your work item to 'review' (NOT 'done') and return AGENT_COMPLETE with a verification field. The validator closes work items — you do not."`**
      - **`"COMPLETION PROOF REQUIRED in AGENT_COMPLETE: list every file written, the exact commit hash, and paste the vitest/tsc output. A report without this evidence will be rejected."`**
      - **`"If you find that a file or feature already exists: you MUST still verify it satisfies the full task spec before marking review. Finding a file is not completion. Run verification, check every requirement, and report what you found vs. what was required."`**
+     - **The decomposition items from the work item's checklist** (all `decomp-*` prefixed items). Include them verbatim in the prompt and instruct the agent:
+       ```
+       DECOMPOSITION CHECKLIST — you MUST implement/verify each row and tick it immediately via update_checklist_item:
+       - decomp-ui-xxx: <route/file> | action=<action> | expect=<result> | evidence=<artifact>
+       - decomp-api-xxx: <route/file> | action=<action> | expect=<result> | evidence=<artifact>
+       Tick each item as soon as that specific behavior is implemented and proven. Do NOT batch.
+       ```
      - **The test plan items from the work item's checklist** (all `test-*` prefixed items). Include them verbatim in the prompt and instruct the agent:
        ```
        TEST PLAN — you MUST implement/verify each of these and tick them off via update_checklist_item:
@@ -263,6 +310,11 @@ Read the task title and description, then:
     - Runs `npx tsc --noEmit` for every touched app/package
     - Starts the dev server and probes every modified route (expects HTTP 200, not 500)
     - Runs vitest for every touched package
+    - **Verifies checklist decomposition quality per work item before functional validation:**
+      - Every implementation work item has at least one `decomp-*` item and one `test-*` item
+      - Every `decomp-*` item includes route/file, action, expected result, and evidence artifact
+      - Any unchecked `decomp-*` item with `required: true` = work item CANNOT be set to `done`
+      - Any coarse or non-falsifiable `decomp-*` item = reopen to `todo` with the specific failure
     - **Verifies test plan completion per work item:**
       - For each `test-assert-*` checklist item: confirm a corresponding vitest test exists and passes
       - For each `test-smoke-*` checklist item: run the command and confirm exit code / HTTP status

package/skills/handoff/SKILL.md

@@ -83,6 +83,27 @@ Template:
 Wave 1 (parallel): Package 1, Package 2
 Wave 2 (after Wave 1): Package 3
 
+## Checklist Decomposition Matrix
+
+| Work item | Atomic deliverable | Surface type | Route or file path | Preconditions / fixture data | Action | Expected result | Verification artifact | Owner package | Status |
+|-----------|-------------------|--------------|--------------------|------------------------------|--------|-----------------|-----------------------|---------------|--------|
+| WP-1 | decomp-<surface>-<slug> | screen | /route | seeded fixture | open / click / submit | observable result | Playwright screenshot / route probe / test name | Package 1 | todo |
+
+Rules:
+- One observable behavior per row.
+- Each row must be independently pass/fail.
+- Each implementation work item must have matching `decomp-*` checklist rows and `test-*` verification rows.
+- UI routes list state rows separately: empty, loading, loaded, error, create/edit/detail/mobile where applicable.
+- API and DB rows include success, failure/guard, and side-effect proof where applicable.
+
+## Checklist Quality Gate
+
+verdict: PASS | FAIL
+failures:
+- <missing/coarse row if any>
+deferred:
+- <explicitly out-of-scope row if any>
+
 ## Definition of Done
 
 - [ ] [specific acceptance criterion]
@@ -114,6 +135,9 @@ Where: <files>
 Agent type: <type>
 Guide: .rdc/guides/<type>.md (fallback: .rdc/guides/<type>.md)
 Design doc: .rdc/plans/<topic-slug>.md (fallback: .rdc/plans/<topic-slug>.md)
+Checklist: include required rows from the plan matrix:
+- decomp-<surface>-<slug>: route/file=<path> | action=<action> | expect=<result> | evidence=<artifact>
+- test-<type>-<slug>: <verification command or artifact>
 Depends on: <other task if applicable>
 Est: <hours>',
   p_parent_id   := '<epic-uuid>'::uuid,
@@ -125,6 +149,8 @@ Est: <hours>',
 );
 ```
 
+Do not create build-ready tasks unless the plan has `## Checklist Decomposition Matrix` and `## Checklist Quality Gate` with `verdict: PASS`. If the matrix fails, save the plan as `Status: Needs checklist repair` and return the failure list instead of creating dispatchable implementation tasks.
+
 ### Step 4 — Register Prototype (if one was built)
 
 ```sql

package/skills/plan/SKILL.md

@@ -52,6 +52,66 @@ description: "No epic exists and you need architecture + task breakdown. Produce
    - Assign an agent type to each work package from the typed dispatch table in rdc:build. Include the guide file path (from `.rdc/guides/`, fallback `.rdc/guides/`) in each work package description.
    - Estimate: small (1 agent, <500 LOC), medium (1 agent, 500-1500 LOC), large (needs splitting)
 
+4b. **Build the checklist decomposition matrix (MANDATORY PRE-BUILD GATE):**
+
+   Before writing Supabase work items, create a `## Checklist Decomposition Matrix`
+   in the plan doc. This matrix is the source of truth for task checklists and
+   build verification.
+
+   Required columns:
+   - Work item ID or placeholder
+   - Atomic deliverable
+   - Surface type: `screen`, `state`, `action`, `api`, `db`, `migration`, `component`, `asset`, `tool`, `test`, `doc`
+   - Route or file path
+   - Preconditions / fixture data
+   - User or agent action
+   - Expected UI/API/DB result
+   - Verification artifact: test name, route probe, Playwright screenshot, SQL query, API response, type-check, migration proof, or CLI transcript
+   - Owner work package
+   - Status
+
+   Atomicity rubric:
+   - One observable behavior per row.
+   - Each row names a concrete route or file path.
+   - Each row names one concrete verification artifact.
+   - Each row can independently pass or fail.
+   - Each row is small enough for a worker to implement and tick without hidden intent.
+
+   Required decomposition by surface:
+   - UI screens: list empty, loading, loaded, error, create, edit, detail, delete/archive guard, mobile, and auth states where applicable.
+   - UI actions: open, search, filter, select, duplicate, save, assign, activate, archive, delete, import, apply, cancel where applicable.
+   - API routes: successful read/write, validation failure, unauthorized/forbidden, and side-effect verification where applicable.
+   - DB/migrations: table/column/index/policy/trigger/function, FK/guard, rollback or smoke query, and type exposure.
+   - CLI/sidebar/local tools: start, attach, enqueue, poll, reply, timeout/not-found, and live refresh where applicable.
+   - Visual work: each named screenshot and visual checkpoint gets its own row.
+   - Cross-system workflows: each handoff boundary gets its own row.
+
+   Minimum row-count heuristics:
+   - UI route: at least 4 rows.
+   - CRUD surface: at least 6 rows.
+   - API route: at least 3 rows.
+   - DB work package: at least 5 rows.
+   - Local editor/sidebar workflow: at least 5 rows.
+
+   Reject these checklist items as too coarse:
+   - "theme management works"
+   - "build all screens"
+   - "verify UI"
+   - "integration complete"
+   - "tests pass"
+
+   Replace them with rows like:
+   - `decomp-ui-theme-manager-loaded: /brands/[id]/theme shows owned theme rows with status, project usage, and actions; evidence: Playwright screenshot`
+   - `decomp-action-duplicate-theme: duplicate submits source brand_theme_id and creates a new editable brand-owned copy; evidence: vitest + DB query`
+   - `decomp-api-import-validation: POST /api/tools/theme-import rejects missing source URL with 400 JSON error; evidence: route probe`
+
+   Add a `## Checklist Quality Gate` section with:
+   - `verdict: PASS` only when every row passes the rubric.
+   - `failures:` list any coarse, missing, duplicate, or unverifiable rows.
+   - `deferred:` list any explicit out-of-scope rows.
+
+   Do not create build-ready work items unless this gate is `PASS`.
+
 5. **Write a test plan for each work package (MANDATORY):**
 
    Every work package MUST have a `test_plan` section with specific, concrete test items. Each item has a type:
@@ -65,6 +125,8 @@ description: "No epic exists and you need architecture + task breakdown. Produce
 
    **Rules for writing test plan items:**
    - Every item must be a specific, falsifiable assertion — not "write tests" or "verify it works"
+   - Every item must map to one or more rows in the Checklist Decomposition Matrix.
+   - A test plan item may summarize multiple checks only when the matrix still keeps those checks as separate atomic rows.
    - New functions/modules MUST have at least one `assert` item
    - API routes MUST have at least one `smoke` item
    - UI pages MUST have at least one `visual` item
@@ -91,6 +153,8 @@ description: "No epic exists and you need architecture + task breakdown. Produce
    ## Goal
    ## Design Decisions
    ## Work Packages (each with test plan)
+   ## Checklist Decomposition Matrix
+   ## Checklist Quality Gate
    ## Sequencing (what can parallelize, what depends on what)
    ## Risks & Mitigations
    ```
@@ -100,12 +164,16 @@ description: "No epic exists and you need architecture + task breakdown. Produce
    - Epic DoD MUST include: `{"id":"test-plan-verified","text":"All test plan items implemented and passing","required":true,"checked":false}`
    - Set `p_definition_of_done` on the epic — child tasks inserted under it will auto-inherit it as their checklist
    - One task per work package via `insert_work_item(p_parent_id := <epic_id>, ...)` — checklist auto-hydrated from epic's DoD
-   - **Additionally, write test plan items as checklist items** on each task, using id format `test-<type>-<slug>`:
+   - **Additionally, write decomposition rows as checklist items** on each task, using id format `decomp-<surface>-<slug>`.
+     The task checklist MUST include both the atomic `decomp-*` rows and the `test-*` verification rows.
+     The `decomp-*` row text must include the route/file, action, expected result, and evidence artifact.
+   - **Write test plan items as checklist items** on each task, using id format `test-<type>-<slug>`:
      ```sql
      SELECT insert_work_item(
        p_parent_id := '<epic_id>',
        p_title := 'WP-2: AST Scanner',
        p_checklist := '[
+         {"id":"decomp-api-scan-success","text":"api: GET /api/layout/scan?dir=apps/studio/src returns 200 JSON with roots[] and warnings[]; evidence: route probe output","required":true,"checked":false},
          {"id":"test-assert-scanner-filters","text":"assert: scanFile returns only container components","required":true,"checked":false},
          {"id":"test-assert-nesting-warn","text":"assert: invalid nesting produces warnings","required":true,"checked":false},
          {"id":"test-smoke-scan-api","text":"smoke: GET /api/layout/scan returns 200","required":true,"checked":false},
@@ -114,7 +182,7 @@ description: "No epic exists and you need architecture + task breakdown. Produce
        ]'::jsonb
      );
      ```
-   - Agents MUST tick each `test-*` checklist item as they implement/verify it (via `update_checklist_item`)
+   - Agents MUST tick each `decomp-*` and `test-*` checklist item as they implement/verify it (via `update_checklist_item`)
    - `update_work_item_status('done')` will REJECT if any `required: true` item is unchecked — this is the enforcement gate
    - Set priorities: urgent/high/normal based on sequencing
 

package/skills/review/SKILL.md

@@ -81,7 +81,8 @@ description: "Post-build quality gate: tsc, tests, stale docs, export conflicts
 
 9. **Verification gate — dispatch the verify agent:**
    After any fixes land, run the verify gate on every touched package. See `guides/agents/verify.md`.
-   **Iron Law: no CLEAN verdict without fresh evidence.** Quote the vitest + tsc output in the report.
+   **Iron Law: no CLEAN verdict without fresh evidence.** Quote the checklist decomposition verdict, vitest output, and tsc output in the report.
+   The verify agent must reject any work item that lacks passed `decomp-*` checklist rows or whose rows are too coarse to prove one observable behavior at a time.
    If verify fails → do NOT emit CLEAN. Loop back, fix, re-run verify.
 
 10. **Fix issues found:**