@lifeaitools/rdc-skills 0.25.1 → 0.25.3

package/RELEASE.md

@@ -0,0 +1,49 @@
+# Release — @lifeaitools/rdc-skills
+
+## Package
+- **Registry:** npm `@lifeaitools/rdc-skills`
+- **Repo:** `LIFEAI/rdc-skills` (standalone, not in regen-root monorepo)
+- **Default branch:** `master`
+
+## Release process
+
+```bash
+# 1. Merge feature branch to master
+git checkout master && git pull origin master
+git merge <feature-branch> --no-edit
+
+# 2. Bump version
+npm version patch|minor|major --no-git-tag-version
+
+# 3. Commit + tag + push
+git add package.json && git commit -m "release: v$(node -e \"console.log(require('./package.json').version)\")"
+git tag "v$(node -e \"console.log(require('./package.json').version)\")"
+git push origin master --tags
+
+# 4. Publish to npm
+npm publish
+
+# 5. Install globally (serves both Claude Code CLI + MCP server)
+npm install -g @lifeaitools/rdc-skills@latest
+
+# 6. Verify
+npm list -g @lifeaitools/rdc-skills --depth=0
+```
+
+## Automated via rdc:release
+
+```
+rdc:release rdc-skills --patch
+```
+
+This runs steps 1–6 automatically.
+
+## Environment targets
+- **Local dev:** `npm install -g` + PM2 `rdc-skills-mcp` process
+- **Remote MCP:** PM2 on Vultr `64.237.54.189` (auto-deploys via webhook on push to master)
+- **Production:** N/A (npm global install is the distribution mechanism)
+
+## Version policy
+- patch: new skill, skill fix, manifest update
+- minor: new capability, breaking skill interface change
+- major: harness architecture change

package/git-sha.json

@@ -1,3 +1,3 @@
 {
-  "sha": "bf46e3a7c1b60f99831e4e087c53007083eac436"
+  "sha": "7a77a9d962212b9731c805820f4f59127fe9ca9b"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lifeaitools/rdc-skills",
-  "version": "0.25.1",
+  "version": "0.25.3",
   "description": "RDC typed-agent dispatch skill suite for Claude Code - plan, build, review, overnight builds",
   "keywords": [
     "claude-code",

package/skills/build/SKILL.md

@@ -281,19 +281,19 @@ Read the task title and description, then:
 
    | Task character | Model | When to pick it |
    |---|---|---|
-   | Updates, edits, mechanical refactors, small fixes, content tweaks, config patches, doc/copy edits, straightforward API wiring | `claude-sonnet-4-6` | Default for `frontend.md`/`content.md`/`infrastructure.md` work whose checklist is mostly "change X to Y" or "wire up endpoint Z". Budget-safe for parallel dispatch. |
-   | Harder coding tasks — non-trivial algorithm, migration with backfill, schema reshape, multi-file refactor with subtle invariants, performance-sensitive code, anything where correctness is the bar | `claude-opus-4-6` | Default for `backend.md`/`data.md` work and any `frontend.md` task that involves state machines, race conditions, or cross-package contracts. |
-   | Design or innovative thought — new component design, brand/UX work, CS 2.0 paradigm work (HAIL/Quad Pixel/AEMG/Virtue), grammar evolution, architecture-first design, anything where the *shape* of the solution is the deliverable rather than the implementation | `claude-opus-4-8` | Default for `design.md`/`cs2.md` work. Also use for `backend.md`/`data.md` tasks tagged with `architecture` or `design-decision` in work item labels. |
+   | Updates, edits, mechanical refactors, small fixes, content tweaks, config patches, doc/copy edits, straightforward API wiring | `claude-sonnet-5` | Default for `frontend.md`/`content.md`/`infrastructure.md` work whose checklist is mostly "change X to Y" or "wire up endpoint Z". Budget-safe for parallel dispatch. |
+   | Harder coding tasks — non-trivial algorithm, migration with backfill, schema reshape, multi-file refactor with subtle invariants, performance-sensitive code, anything where correctness is the bar | `claude-opus-5` | Default for `backend.md`/`data.md` work and any `frontend.md` task that involves state machines, race conditions, or cross-package contracts. |
+   | Design or innovative thought — new component design, brand/UX work, CS 2.0 paradigm work (HAIL/Quad Pixel/AEMG/Virtue), grammar evolution, architecture-first design, anything where the *shape* of the solution is the deliverable rather than the implementation | `claude-opus-5` | Default for `design.md`/`cs2.md` work. Also use for `backend.md`/`data.md` tasks tagged with `architecture` or `design-decision` in work item labels. |
 
    **How to choose when the task straddles categories:**
-   - If the task's checklist contains the word "design", "decide", "propose", "evaluate alternatives", "novel", or any CS 2.0 primitive → **Opus 4.8**.
-   - If the task touches `packages/cs2*`, `packages/hail`, `packages/quad-pixel`, `packages/virtue-engine`, `packages/aemg`, `packages/planetary-ontology`, or `packages/being-state-processor` → **Opus 4.8** (CS 2.0 paradigm requires innovative thought, not transcription).
-   - If the task is a Supabase migration that drops/renames/reshapes anything, or a refactor across ≥5 files → **Opus 4.6**.
-   - Otherwise → **Sonnet 4.6**.
+   - If the task's checklist contains the word "design", "decide", "propose", "evaluate alternatives", "novel", or any CS 2.0 primitive → **Opus 5**.
+   - If the task touches `packages/cs2*`, `packages/hail`, `packages/quad-pixel`, `packages/virtue-engine`, `packages/aemg`, `packages/planetary-ontology`, or `packages/being-state-processor` → **Opus 5** (CS 2.0 paradigm requires innovative thought, not transcription).
+   - If the task is a Supabase migration that drops/renames/reshapes anything, or a refactor across ≥5 files → **Opus 5**.
+   - Otherwise → **Sonnet 5**.
 
    **State the choice in the wave plan.** Before dispatching a wave, the supervisor must log one line per agent in the form `[wave-N agent-K] role=<role> task=<id> model=<chosen> reason=<one phrase>`. This keeps routing decisions reviewable in the transcript and lets `rdc:report` summarize the fleet mix.
 
-   **Cost guardrail.** If a single wave would dispatch more than 3 Opus 4.8 agents in parallel, downshift the lowest-priority Opus-4.8 tasks to Opus 4.6 unless their work items are tagged `priority=urgent`. Opus 4.8 fast-mode is cheap individually but still ~5× a Sonnet agent at scale.
+   **Cost guardrail.** If a single wave would dispatch more than 3 Opus 5 agents in parallel, downshift the lowest-priority Opus 5 tasks to Sonnet 5 unless their work items are tagged `priority=urgent`. Opus 5 is cheap individually but still ~2× a Sonnet 5 agent at scale.
    Without `max_turns: 70`, agents hit the default turn cap mid-task and stop.
    `isolation: "worktree"` gives each agent its own git worktree and branch — eliminates push race conditions and index lock contention when multiple agents commit in parallel. The supervisor merges worktree branches after each wave (Step 9).
 
@@ -570,7 +570,7 @@ NEVER run pnpm build or pnpm turbo. Use npx vitest run only.
 - Push after each wave, not just at the end
 - Unattended: NEVER pause — continue automatically
 - Unattended: max 2 retries per task before escalating to advisor
-- Every Agent() dispatch: `model: <routed>` + `max_turns: 70` — non-negotiable. `isolation` is per the dispatch-mode rule in Step 7: NON-isolated (omit `isolation`) for doc-only and single-committer waves; `isolation: "worktree"` ONLY for true parallel MULTI-committer waves, and only after the HARD GATE (worktree base == develop HEAD) passes — a stale-base isolated wave is a build failure. Model is chosen per task per the routing table in Step 7: Sonnet 4.6 for updates/edits, Opus 4.6 for harder coding, Opus 4.8 for design/innovative thought (CS 2.0, brand/UX, architecture). Supervisor logs `model=<chosen> reason=<phrase>` per agent. Exception: validator agent in Step 10 always omits isolation; validator model stays `claude-sonnet-4-6` (verification, not generation).
+- Every Agent() dispatch: `model: <routed>` + `max_turns: 70` — non-negotiable. `isolation` is per the dispatch-mode rule in Step 7: NON-isolated (omit `isolation`) for doc-only and single-committer waves; `isolation: "worktree"` ONLY for true parallel MULTI-committer waves, and only after the HARD GATE (worktree base == develop HEAD) passes — a stale-base isolated wave is a build failure. Model is chosen per task per the routing table in Step 7: Sonnet 5 for updates/edits, Opus 5 for harder coding and design/innovative thought (CS 2.0, brand/UX, architecture). Supervisor logs `model=<chosen> reason=<phrase>` per agent. Exception: validator agent in Step 10 always omits isolation; validator model stays `claude-sonnet-5` (verification, not generation).
 - Finding an existing file is NOT task completion — verify it satisfies the spec
 
 ## Capture lessons (exit step)

package/skills/env/SKILL.md

@@ -77,14 +77,12 @@ Check each service:
 | Service | Health Check | Repair |
 |---------|-------------|--------|
 | clauth daemon | `curl -s http://127.0.0.1:52437/ping` | `$LIFEAI_ENV/services/restart-clauth.bat` |
-| CodeFlow gateway | `curl -s http://127.0.0.1:3109/health` | `node $PROJECT_ROOT/scripts/codeflow-up.mjs` |
-| CodeFlow brain | `/health` → `health.state` should be `operational` | `node $PROJECT_ROOT/scripts/codeflow-up.mjs --restart` |
-| rdc-skills MCP | `pm2 list` shows rdc-skills-mcp online | `pm2 restart rdc-skills-mcp` |
+| CodeFlow gateway | `curl -s http://127.0.0.1:3109/health` | **PROBE ONLY — never start/restart/recreate.** CodeFlow is blue-green-owned. If down, report `BLOCKED: CodeFlow gateway down — run: node scripts/codeflow-bluegreen.mjs recover` |
+| CodeFlow brain | `/health` → `health.state` | **REPORT ONLY.** `degraded` = freshness drift (normal after pushes). `offline` = remote PM2 brain unreachable. Never start Docker Neo4j or local brain from this skill. |
+| rdc-skills MCP | `pm2 list` shows rdc-skills-mcp online | `npm install -g @lifeaitools/rdc-skills@latest` (reinstall picks up latest) |
 | PM2 daemon | `pm2 ping` | `pm2 resurrect` |
-| Docker | `docker info` | Start Docker Desktop |
-| Neo4j | `docker inspect codeflow-neo4j` | `docker compose -f $LIFEAI_ENV/services/codeflow/docker-compose.yml up -d` |
 
-For `status`: report only. For `repair`: fix each failing service in order (clauth first, then CodeFlow, then MCPs).
+For `status`: report only. For `repair`: fix clauth + rdc-skills only. CodeFlow repair is exclusively through `codeflow-bluegreen.mjs recover` — this skill never touches CodeFlow lifecycle, Docker, or Neo4j.
 
 ### Step 4: MCP server verification
 

package/skills/housekeeping/SKILL.md

@@ -174,7 +174,7 @@ For each cluster:
 
 When the weekly triage runs unattended, follow `.rdc/guides/lessons-learned-spec.md` § **Triage procedure — UNATTENDED weekly mode** — do not run `AskUserQuestion`. In brief (the spec is authoritative):
 
-- **Per-difficulty model routing** (reuses the `rdc:build` table): the run is led by `claude-opus-4-8` for clustering + scope/difficulty triage; mechanical apply → `claude-sonnet-4-6`; harder multi-file/migration fix → `claude-opus-4-6`; design/architectural fix → `claude-opus-4-8`.
+- **Per-difficulty model routing** (reuses the `rdc:build` table): the run is led by `claude-sonnet-5` for clustering + scope/difficulty triage; mechanical apply → `claude-haiku-4-5`; harder multi-file/migration fix → `claude-sonnet-5`; design/architectural fix → `claude-opus-5`.
 - `scope: simple` → apply directly or via `rdc:fixit`, commit, `status: applied`.
 - `scope: architectural` with a single clear correct fix (records an already-learned lesson — e.g. "document X", "add a gate", update a guide) → route through `rdc:plan` → `rdc:build` (or `rdc:fixit` if genuinely <5 files), update the documentation, set `status: applied`/`triaged`.
 - `scope: architectural` **and genuinely ambiguous** (multiple valid approaches, real tradeoffs) → write a `human_items` row (`item_type='decision'`, with options in `suggested_agent_prompt`, `source_type='lesson'`, `source_fingerprint` for dedupe), set `status: triaged`, spawn + link a `work_item`. Decided in the morning. This is the asynchronous equivalent of the attended interview and honors `.claude/rules/architectural-change-approval.md`.