@mmerterden/multi-agent-pipeline 20.36.0 → 20.37.0

package/pipeline/skills/shared/core/google-play-compliance/SKILL.md

@@ -271,80 +271,18 @@ On Android diffs, reviewer agent annotates new/changed permissions with their po
 
 ## Examples
 
-### Primary: `/multi-agent:test "store-ready"`
-
-```
-# Full report on most recent bundle (auto-pick)
-/multi-agent:test "store-ready" --auto-pick
-
-# Specific .aab, blockers only, Turkish
-/multi-agent:test "store-ready" app/build/outputs/bundle/release/app-release.aab --threshold=error --lang=tr
-
-# Supply last version code for progression check
-multi-agent-test "store-ready" --last-version-code=142 --auto-pick
-
-# Raw JSON for CI
-/multi-agent:test "store-ready" --format=json --auto-pick
-```
-
-### Phase 3 Security Auditor (catalog cross-ref, no tool invocation)
-
-When a Phase 3 review diff touches Android release-relevant paths (`AndroidManifest.xml`, `build.gradle(.kts)`, `proguard-rules.pro`, `network_security_config.xml`), the security-auditor persona loads this skill's 21-rule policy catalog and cites matching ruleIDs next to its findings. No bundletool / aapt2 / apksigner invocation at review time  -  catalog reads are free.
-
-```
-# Example finding from Phase 3 Security Auditor (wired in v5.8.1):
-[BLOCKING] AndroidManifest.xml: new MANAGE_EXTERNAL_STORAGE permission added
-  (google-play-compliance / dangerous-permissions  -  Policy  -  Permissions)
-  File: app/src/main/AndroidManifest.xml:12
-  Fix: Remove the permission or add Play Console justification before submitting.
-       MANAGE_EXTERNAL_STORAGE triggers mandatory policy review on every update.
-```
-
-### `/multi-agent:review` (PR citation)
-
-```
-# Standard review  -  auto-loads Google catalog when diff touches Android release files
-/multi-agent:review
-
-# Review a specific branch
-/multi-agent:review feature/PROJ-67890-file-manager
-```
-
-Findings in the review report carry the same `(google-play-compliance / <ruleID>  -  <Play ref>)` format as Phase 4  -  the review command reuses the catalog contract.
-
-### `/multi-agent:channels` (PR body auto-augmentation)
-
-```
-# Run the compliance scan first so its JSON lands in /tmp
-/multi-agent:test "store-ready" --auto-pick --last-version-code=142
-
-# Then deliver the report  -  channels auto-appends a Store compliance section
-#   to the PR body / Jira comment / Confluence page if the cache is < 6h old
-#   and at least one blocker or warning was found.
-/multi-agent:channels PROJ-67890 --channels pr,jira
-```
-
-The appended section is structured:
-
-```markdown
-### Store compliance
-
-⚠️ 3 blocker(s), 2 warning(s)  -  run /multi-agent:test "store-ready" for the full report.
-
-- `target-api-level`  -  Policy  -  Target API: targetSdkVersion = 35 but Play Console requires ≥ 36 for new apps and updates
-- `native-64bit`  -  Policy  -  64-bit: lib/armeabi-v7a/libfoo.so has no arm64-v8a counterpart
-- `foreground-service-types`  -  Android 14: <service> missing android:foregroundServiceType
-+ 2 more
-```
+Invocation examples for each consumer (`/multi-agent:test "store-ready"`, the Phase 3 Security Auditor
+catalog cross-reference, `/multi-agent:review` citations and the `/multi-agent:channels` Store compliance
+section) are in [references/examples.md](references/examples.md).
 
 ## Notes
 
-- **Policy thresholds change yearly.** The target API level in Rule #1 advances each August for new apps / November for updates. When a new Android version ships, update Rule #1 `target-api-level` with the new threshold.
-- **Not a substitute for Play Console pre-launch report.** This skill catches static issues (manifest, signing, size, debug tools). Dynamic issues (crashes, ANRs, security vulnerabilities) require Play Console's own pre-launch testing on real devices.
-- **Data Safety form audit is partial.** Rule #14 flags manifest permissions that commonly trigger form updates, but the declared answers live in Play Console and cannot be cross-checked from here. The Android Publisher API has no Data Safety read: its only operation, `applications.dataSafety`, writes the declaration. With `google_play` onboarded, the toolkit's `store_search` / `store_call` reach the rest of the Android Publisher and Play Developer Reporting APIs (tracks, bundles, vitals), but the gap for Data Safety stays.
+The target API threshold in Rule #1 moves every year, the scan is static and does not replace the Play Console
+pre-launch report, and the Data Safety check is partial: [references/notes.md](references/notes.md).
 
-## Version history
+## References
 
-- **v5.8.0**  -  Skill introduced. Orchestrates bundletool + aapt2 + apksigner, 21-rule catalog, EN+TR humanizer. Primary consumer only (`/multi-agent:test "store-ready"`).
-- **v5.8.1**  -  3 additional consumer surfaces wired (Phase 3 Security Auditor cross-ref, `/multi-agent:review` citation, `/multi-agent:channels` PR-body auto-augmentation). Regression-locked by `smoke-compliance-skills.sh` (45 assertions).
-- **v6.0.0**  -  `language: en` frontmatter tag; 4-consumer Examples section expanded.
+- [rules.yml](references/rules.yml): the machine-readable rule catalog behind the catalog table.
+- [Examples](references/examples.md): open for command lines, a sample auditor finding and the exact Store compliance section channels appends.
+- [Limits and caveats](references/notes.md): open before trusting a clean result or when a new Android version ships.
+- [Version history](references/version-history.md): which release introduced each consumer surface.

package/pipeline/skills/shared/core/google-play-compliance/references/examples.md

@@ -0,0 +1,69 @@
+# Google Play compliance: Examples
+
+Invocation examples for each consumer surface of the skill.
+
+## Primary: `/multi-agent:test "store-ready"`
+
+```
+# Full report on most recent bundle (auto-pick)
+/multi-agent:test "store-ready" --auto-pick
+
+# Specific .aab, blockers only, Turkish
+/multi-agent:test "store-ready" app/build/outputs/bundle/release/app-release.aab --threshold=error --lang=tr
+
+# Supply last version code for progression check
+multi-agent-test "store-ready" --last-version-code=142 --auto-pick
+
+# Raw JSON for CI
+/multi-agent:test "store-ready" --format=json --auto-pick
+```
+
+## Phase 3 Security Auditor (catalog cross-ref, no tool invocation)
+
+When a Phase 3 review diff touches Android release-relevant paths (`AndroidManifest.xml`, `build.gradle(.kts)`, `proguard-rules.pro`, `network_security_config.xml`), the security-auditor persona loads this skill's 21-rule policy catalog and cites matching ruleIDs next to its findings. No bundletool / aapt2 / apksigner invocation at review time  -  catalog reads are free.
+
+```
+# Example finding from Phase 3 Security Auditor (wired in v5.8.1):
+[BLOCKING] AndroidManifest.xml: new MANAGE_EXTERNAL_STORAGE permission added
+  (google-play-compliance / dangerous-permissions  -  Policy  -  Permissions)
+  File: app/src/main/AndroidManifest.xml:12
+  Fix: Remove the permission or add Play Console justification before submitting.
+       MANAGE_EXTERNAL_STORAGE triggers mandatory policy review on every update.
+```
+
+## `/multi-agent:review` (PR citation)
+
+```
+# Standard review  -  auto-loads Google catalog when diff touches Android release files
+/multi-agent:review
+
+# Review a specific branch
+/multi-agent:review feature/PROJ-67890-file-manager
+```
+
+Findings in the review report carry the same `(google-play-compliance / <ruleID>  -  <Play ref>)` format as Phase 4  -  the review command reuses the catalog contract.
+
+## `/multi-agent:channels` (PR body auto-augmentation)
+
+```
+# Run the compliance scan first so its JSON lands in /tmp
+/multi-agent:test "store-ready" --auto-pick --last-version-code=142
+
+# Then deliver the report  -  channels auto-appends a Store compliance section
+#   to the PR body / Jira comment / Confluence page if the cache is < 6h old
+#   and at least one blocker or warning was found.
+/multi-agent:channels PROJ-67890 --channels pr,jira
+```
+
+The appended section is structured:
+
+```markdown
+### Store compliance
+
+⚠️ 3 blocker(s), 2 warning(s)  -  run /multi-agent:test "store-ready" for the full report.
+
+- `target-api-level`  -  Policy  -  Target API: targetSdkVersion = 35 but Play Console requires ≥ 36 for new apps and updates
+- `native-64bit`  -  Policy  -  64-bit: lib/armeabi-v7a/libfoo.so has no arm64-v8a counterpart
+- `foreground-service-types`  -  Android 14: <service> missing android:foregroundServiceType
++ 2 more
+```

package/pipeline/skills/shared/core/google-play-compliance/references/notes.md

@@ -0,0 +1,7 @@
+# Google Play compliance: Limits and Caveats
+
+What the static scan does not cover and which thresholds move over time.
+
+- **Policy thresholds change yearly.** The target API level in Rule #1 advances each August for new apps / November for updates. When a new Android version ships, update Rule #1 `target-api-level` with the new threshold.
+- **Not a substitute for Play Console pre-launch report.** This skill catches static issues (manifest, signing, size, debug tools). Dynamic issues (crashes, ANRs, security vulnerabilities) require Play Console's own pre-launch testing on real devices.
+- **Data Safety form audit is partial.** Rule #14 flags manifest permissions that commonly trigger form updates, but the declared answers live in Play Console and cannot be cross-checked from here. The Android Publisher API has no Data Safety read: its only operation, `applications.dataSafety`, writes the declaration. With `google_play` onboarded, the toolkit's `store_search` / `store_call` reach the rest of the Android Publisher and Play Developer Reporting APIs (tracks, bundles, vitals), but the gap for Data Safety stays.

package/pipeline/skills/shared/core/google-play-compliance/references/version-history.md

@@ -0,0 +1,5 @@
+# Google Play compliance: Version History
+
+- **v5.8.0**  -  Skill introduced. Orchestrates bundletool + aapt2 + apksigner, 21-rule catalog, EN+TR humanizer. Primary consumer only (`/multi-agent:test "store-ready"`).
+- **v5.8.1**  -  3 additional consumer surfaces wired (Phase 3 Security Auditor cross-ref, `/multi-agent:review` citation, `/multi-agent:channels` PR-body auto-augmentation). Regression-locked by `smoke-compliance-skills.sh` (45 assertions).
+- **v6.0.0**  -  `language: en` frontmatter tag; 4-consumer Examples section expanded.

package/pipeline/skills/shared/core/multi-agent/SKILL.md

@@ -13,307 +13,84 @@ Read and follow the instructions in this file's INSTRUCTION.md section below.
 
 # Multi-Agent Task Orchestrator
 
-References under `$HOME/.claude/multi-agent-refs/` are installed with Claude Code.
-
-## Project Detection
-
-Detect project from cwd FIRST, then validate input format:
-
-| cwd contains | Project | Accepted Input | Rejected Input |
-|--------------|---------|----------------|----------------|
-| `/my-ios-app` | my-app | `"PROJ-XXX" "branch"` | GitHub Issue URL |
-| `/my-figma-app` | figma | `"https://github.com/.../issues/N"` | Jira ID + branch |
-| Other | generic | Both formats |  -  |
-
-If user gives wrong format for the project, warn:
-- my-ios-app + GitHub URL → "This project expects a Jira ID + branch. Example: multi-agent \"PROJ-12345\" \"feature/PROJ-12345-desc\""
-- my-figma-app + Jira → "This project expects a GitHub Issue URL. Example: multi-agent \"https://github.com/my-org/my-figma-app/issues/316\""
-
-## Input Parsing
-
-Classify user input into one of 5 types (matches the `commands/multi-agent/SKILL.md` dispatcher contract; cross-CLI parity is enforced by `smoke-cross-cli-behavior.sh`):
-
-| # | Pattern | Type | Action |
-|---|---------|------|--------|
-| 1 | `#N` or bare `316` | github-issue-number | Fetch via gh after project selection |
-| 2 | `https://github.com/.../issues/N` | github-issue-url | Extract org/repo/N, fetch via gh |
-| 3 | `https://{JIRA_HOST}/browse/{JIRA_KEY}-XXXXX` | jira-url | Extract ID, fetch via Jira API |
-| 4 | `{JIRA_KEY}-XXXXX` (bare) | jira-id | Fetch via Jira API |
-| 5 | Anything else | free-text | No external fetch |
-
-All types enter the same Phase 0 flow.
+References under `$HOME/.claude/multi-agent-refs/` are installed with Claude Code. Before routing, read `$HOME/.claude/multi-agent-refs/orchestrator/copilot-pipeline.md`: project detection (cwd decides which input format is accepted, and the wrong format gets a warning), the five input types, the full routing rows, the autopilot matrix, the persona models and the per-phase summaries.
 
 ## Routing
 
-Parse the user's input and route to the correct sub-command:
-
-| Input Pattern | Action |
-|---------------|--------|
-| `"PROJ-XXX" "branch"` | **my-app only**  -  Jira ID + branch name → auto-assigned incremental ID |
-| `"https://github.com/..."` | **figma only**  -  parse issue, extract Jira from title/body, auto-create branch |
-| `status` | Show all tasks with ID, phase, status |
-| `log [id]` | Show agent-log.md for task (latest if no id) |
-| `resume [id]` | Resume paused/failed task from last successful phase |
-| `kill [id]` | Stop task, delete worktree + branch + logs. Asks confirm first |
-| `clear-logs` | Superseded  -  redirect only, see below |
-| `purge` | Nuclear cleanup  -  delete ALL worktrees, branches, logs, state, counter. Full reset |
-| `review` | Skip phases 0-3, review current diff only |
-| `autopilot` flag | Add to any pipeline command  -  skip user confirmations except destructive ops |
-| No args / `help` | Show usage guide (project-specific) |
+`status`, `log [id]`, `resume [id]`, `kill [id]`, `clear-logs`, `purge`, `review` (current diff only), the `autopilot` flag, and no args / `help`. Anything else is a task input (Jira ID + branch, GitHub issue URL or number, free text) and enters Phase 0.
 
-## Task operations: status, kill, purge, resume
-
-Every task gets an auto-incremented short ID (`#1`, `#2`), stored in `agent-state.json`.
+## Task operations
 
 - **kill** MUST ask for confirmation before deleting anything; the remote branch is deleted ONLY on a second, explicit yes.
 - **purge** MUST confirm twice: a first confirm, then a native `AskUserQuestion` paranoia gate where anything other than **Purge permanently** cancels. Remote branches are deleted ONLY on explicit confirmation.
-- **resume** reads `$HOME/.claude/logs/multi-agent/{project}/{task-id}/agent-state.json` (or `.../artifacts/agent-state.json` for a task Phase 4 finalized) and restarts from `currentPhase + 1`.
-- **clear-logs**: **do not run it.** Say it was superseded, then route to `multi-agent-prune-logs` (per-task logs, audit trail and metrics kept) or `multi-agent-garbage-collect` (`/tmp` scratch + worktree residue). Logs live at `$HOME/.claude/logs/multi-agent/{project}/{task-id}/`, not under `.worktrees/`.
-
-Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/operations.md`. Read it when the input is `status`, `kill`, `purge` or `resume`.
-
----
-
-## Autopilot Mode
+- **resume** reads `$HOME/.claude/logs/multi-agent/{project}/{task-id}/agent-state.json` (or `.../artifacts/agent-state.json`) and restarts from `currentPhase + 1`.
+- **clear-logs**: **do not run it.** Say it was superseded and route to `multi-agent-prune-logs` or `multi-agent-garbage-collect`.
 
-Add `autopilot` to any pipeline command (`multi-agent "PROJ-12345" "feature/PROJ-12345-desc" autopilot`) to skip interactive confirmations and run end-to-end.
+Read `$HOME/.claude/multi-agent-refs/orchestrator/operations.md` when the input is `status`, `kill`, `purge` or `resume`.
 
-| Phase | Normal | Autopilot |
-|-------|--------|-----------|
-| Phase 1 (Plan Approval Gate) | Clarification (max 2 rounds) + approval loop  -  user replies `onayla`/`iptal`/free-text | **Gate skip**  -  log the plan, go straight to Phase 2 (autopilot contract: zero interaction) |
-| Phase 3 (User test) | "Want to test it?" → wait | Skip → go straight to Phase 4 |
-| Phase 4 (Commit) | "Want to commit?" → wait | Automatic commit + push |
-| Phase 4 (PR) | "Want to open a PR?" → wait | Automatically create the PR |
+## Autopilot
 
-**What NEVER skips (even in autopilot):**
-- Phase 3 Review → if there is a blocking finding, return to Phase 2, fix + rebuild automatically (safety)
-- Kill/Purge confirmations → destructive operations always ask
-- Build fail → fix + rebuild automatically (max 3 retries). If it still fails after 3 retries → pause, ask the user
+`autopilot` skips confirmations: the Plan Approval Gate becomes a **Gate skip** (log the plan, go to Phase 2), the user test is skipped, commit, push and PR run automatically. **Never skipped:** a blocking review finding returns to Phase 2; kill / purge confirmations always ask; a build failure retries at most 3 times, then pauses and asks. Read `$HOME/.claude/multi-agent-refs/phases/modes.md` when `autopilot` is set.
 
-**State tracking**: `"autopilot": true` is added to `agent-state.json`. Autopilot continues on resume too.
+## Agent dispatch - per-persona model routing
 
-Full procedure: `$HOME/.claude/multi-agent-refs/phases/modes.md`. Read it when `autopilot` is set, for the circuit-breaker halts and the Phase 5 channels pause.
-
----
-
-## Agent dispatch  -  per-persona model routing
-
-Every agent persona under `$HOME/.claude/agents/*.md` declares a `preferredModel` frontmatter field. The orchestrator MUST honor it on every dispatch, on both Claude Code and Copilot CLI  -  cross-CLI parity is enforced by `smoke-cross-cli-behavior.sh`. Each persona file carries its own `modelRationale`; this is the rung only.
-
-| Persona | `preferredModel` |
-|---|---|
-| `explorer` | `sonnet` |
-| `code-reviewer` | `fable` (Reviewer 1; Phase 3 sets Reviewer 2 `opus`, Reviewer 3 `sonnet`) |
-| `security-auditor` | `opus` |
-| `ios-architect` / `android-architect` / `backend-architect` | `fable` |
-
-**Dispatch contract:**
+Every persona under `$HOME/.claude/agents/*.md` declares `preferredModel`; honor it on every dispatch, on both CLIs.
 
 ```bash
 persona_file="$HOME/.claude/agents/${persona}.md"
 pref_model="$(awk '/^preferredModel:/ {print $2; exit}' "$persona_file")"
-override="${PHASE_MODEL_OVERRIDE:-}"   # per-dispatch, e.g. Phase 3 Reviewer 3 → sonnet
+override="${PHASE_MODEL_OVERRIDE:-}"
 routed="${override:-$("$HOME/.claude/lib/model-dispatch.sh" subagent --persona "$persona" --phase "$currentPhase" --task-kind "$taskType" --default "${pref_model:-opus}" 2>/dev/null)}"
 resolved_model="${routed:-${pref_model:-opus}}"
-
-export CLAUDE_CODE_SUBAGENT_MODEL="$resolved_model"
-# Claude Code honors the env var; Copilot CLI also passes it as --model.
+export CLAUDE_CODE_SUBAGENT_MODEL="$resolved_model"   # Copilot CLI also passes it as --model
 ```
 
-**Precedence:** per-dispatch `PHASE_MODEL_OVERRIDE` > `prefs.global.modelRouting` rule > persona `preferredModel` > global default (`opus`). Routing here is Anthropic-only: features/model-fallback.md "Routing at dispatch".
-
-**Simple text tasks (`delegate`):** before summarizing a log or document, extracting fields, labelling or drafting a translation or commit body in-session, ask `"$HOME/.claude/lib/model-dispatch.sh" delegate --default host`. `free` means call `llm_delegate`; `ESCALATE_TO_CLAUDE` or any other answer means do it in-session. Never for code, plans, reviews or triage. Task list and rules: `$HOME/.claude/multi-agent-refs/features/free-models.md`.
-
-**Observability:** emit `phase.agent.dispatch` OTel span with `{persona, resolved_model, override_source}` attributes when `MULTI_AGENT_OTEL_SPANS=1` (see `phase-tracker.sh`).
-
-## Skill loading and injection
-
-Gated by `prefs.global.dynamicSkillLoading` (default `false`): when on, only skills matched by `match-skills.mjs` are injected into Phase 1-3 sub-agent prompts; without `.skills-index.json` it eager-loads, so it never blocks the happy path. Every sub-agent prompt SHOULD carry task-specific context from Phase 1 on top of the persona's embedded skills.
-
-Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/skill-loading.md`. Read it when `dynamicSkillLoading` is on, or when composing a sub-agent prompt.
+**Precedence:** per-dispatch `PHASE_MODEL_OVERRIDE` > `prefs.global.modelRouting` rule > persona `preferredModel` > global default (`opus`). Anthropic-only: features/model-fallback.md "Routing at dispatch". Simple text tasks ask `model-dispatch.sh delegate --default host` first; `free` means `llm_delegate`, never for code, plans, reviews or triage (`$HOME/.claude/multi-agent-refs/features/free-models.md`).
 
----
+Skill injection (`prefs.global.dynamicSkillLoading`): read `$HOME/.claude/multi-agent-refs/orchestrator/skill-loading.md` when it is on or when composing a sub-agent prompt.
 
 ## Phase Pipeline
 
-Run phases sequentially. Log every step to `agent-log.md`.
-If any phase fails, mark task as `paused` and stop  -  user can `resume` later.
-
-**The blocks below are the routing summary, not the specification.** Each phase's
-full contract lives in `$HOME/.claude/multi-agent-refs/phases/phase-{n}-{name}.md`
-and is read when that phase actually starts. Modes are in `phases/modes.md`,
-operations in `phases/operations.md`, the log shape in `phases/log-format.md`.
-Where a summary here and a ref disagree, the ref wins.
+Run phases sequentially, log every step to `agent-log.md`, and on a failure mark the task `paused` and stop. Each phase's contract is `$HOME/.claude/multi-agent-refs/phases/phase-N-<name>.md`, read when that phase starts; the ref wins over any summary. Update `agent-state.json` at EVERY phase transition.
 
 ### Canonical phase labels (TaskCreate + banner)
 
-Every user-facing phase title  -  TaskCreate cards, `phase-banner.sh` headers, and any status line rendered to the terminal  -  MUST come from this table. Phase labels are English on every host (`promptLanguage` is fixed to `en`; picker text follows its own matrix):
-
-| # | Label                |
-|---|----------------------|
-| 0 | Phase 0: Init        |
-| 1 | Phase 1: Plan        |
-| 2 | Phase 2: Dev         |
-| 3 | Phase 3: Review      |
+| # | Label |
+|---|---|
+| 0 | Phase 0: Init |
+| 1 | Phase 1: Plan |
+| 2 | Phase 2: Dev |
+| 3 | Phase 3: Review |
 | 4 | Phase 4: Commit & PR |
-| 5 | Phase 5: Report      |
+| 5 | Phase 5: Report |
 
-**Render rule:**
-
-1. For TaskCreate, use the matching label as the task `subject`.
-2. For terminal banners, call `phase-banner.sh` with the `auto` sentinel and `PHASE_LANG=en` env  -  or pass the literal label from the table. Never invent a new label.
-3. Reviewer-count qualifiers (e.g. `(parallel + triage)`) stay English  -  they are technical artifacts, not user copy.
-
-**Out of scope for localization** (always EN, per `prefs.schema.json`): reviewer/triage system prompts, commit messages, PR titles/bodies, Jira comments, wiki content, log payloads inside `agent-log.md`  -  these are external artifacts consumed by other tools and humans beyond this user.
+TaskCreate subjects and `phase-banner.sh` headers (`auto`, `PHASE_LANG=en`) use these labels only, in English on every host. Never invent a label.
 
 #### TaskCreate ordering (strict)
 
-All TaskCreate calls fire in strict phase-number order BEFORE any TaskUpdate is applied (Claude Code path; Copilot CLI has no TaskList widget, so this rule is Claude-only). The widget renders by creation order, so out-of-order calls scramble the tile stack even when the state file is correct. Pre-marking phases as completed/skipped before Phase 0 starts is FORBIDDEN: register each tile in order as `pending`, then flip via TaskUpdate when the phase actually short-circuits. Mode-specific phase sets:
-
-- `multi-agent`, `multi-agent-autopilot`: 0 → 1 → 2 → 3 → 4 → 5
-- `multi-agent-analysis`: 0 → 1 → 3 → 4 → 5 (no code is written, so Dev is not in the set)
-
-Every mode registers its whole set at Step -1. Full contract: `refs/tracker-contract.md` section "TaskCreate ordering (strict)".
+All TaskCreate calls fire in strict phase-number order BEFORE any TaskUpdate (Claude Code only; Copilot CLI has no widget). Never pre-mark a phase completed or skipped: register every tile as `pending`, then flip it. `multi-agent`, `multi-agent-autopilot`: 0-5; `multi-agent-analysis`: 0, 1, 3, 4, 5. Contract: `refs/tracker-contract.md`.
 
 ### Phase 0: Init
 
-> **Contract (strict)**  -  Phase 0 has **8 sequential interactive steps**, defined in
-> `$HOME/.claude/multi-agent-refs/phases/phase-0-init.md`. Read it before every Phase 0 run and execute all of
-> them. Do **NOT** short-circuit: even when the input is a Jira ID plus a branch, the
-> prefs load, the multi-repo offer, the branch picker and the branch-name confirm are
-> required UX checkpoints.
->
-> **Exit gate (BLOCKING).** Before marking Phase 0 completed:
->
-> ```bash
-> node "$HOME/.copilot/scripts/phase0-exit-gate.mjs" "$TASK_ID" --input "$USER_INPUT"
-> ```
->
-> Non-zero means Phase 0 did not produce its own output - `taskType`, `baseBranch`,
-> `baseFetchStatus`, a worktree path distinct from the project root, and a recorded Figma
-> access tier when the input carries a Figma URL. Fix the missing step; do not proceed.
->
-> **Credential inventory (Step 0).** Run
-> `bash "$HOME/.copilot/lib/credential-inventory.sh" --json` and store the result in
-> `agent-state.json.credentialInventory`. Per `refs/keychain.md` Rule 2, never ask the
-> user for data a mapped credential can fetch. Ask for the pointer (the issue URL)
-> instead, and say what you will do with it.
-
-Per-project input parsing (my-ios-app / my-figma-app / generic), worktree creation, git author and the initial log + state files. Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/phase-0-projects.md`. Read it when Phase 0 reaches input parsing and worktree setup.
-
-**IMPORTANT**: Update `agent-state.json` at EVERY phase transition. This file is the resume source of truth.
-
-**Issue family** (Jira input, once `agent-state.json` exists): run
-`bash "$HOME/.copilot/lib/jira-family-write.sh" resolve --state "$STATE_FILE"`. It records the
-story, dev and test sub-tasks and the story format check in `state.jiraFamily`, and asks and writes nothing;
-the channels Jira step writes to them.
-
-### Phase 1: Plan
-Two halves, one phase, one approval gate.
-
-**1a  -  Analysis** (claude-sonnet-5-5, `explorer` persona)
-1. Launch **explore agents** (parallel) to scan the codebase: related files, existing patterns and conventions, potential impact areas
-2. Summarize findings
-3. Log: `📊 Phase 1: Plan (analysis)  -  {N} files identified, {summary}`
-
-**1b  -  Planning** (claude-fable-5-1)
-1. Create task breakdown → todos with dependencies
-2. Launch **ios-architect** agent for architecture review (if structural changes)
-3. Determine development approach per todo (new file, modify, refactor)
-4. Log: `🧠 Phase 1: Plan  -  {N} todos created`
-5. **Plan Approval Gate** (interactive only  -  autopilot may not ask). Full flow in `refs/phases/phase-1-plan.md` Step 5:
-   - **5a  -  Clarification** (conditional, max 2 rounds): if the Jira/issue description is ambiguous (vague acceptance, UI task without Figma, API task without endpoint contract, `ambiguityScore >= 2`, parent-story scope drift), ask structured questions before rendering the plan. User answers → plan regenerated. If it is still unclear after the 2nd round, render the plan with a "best-effort" banner.
-   - **5b  -  Approval loop**: render plan → user: `onayla`/`iptal`/free-text. A free-text edit request → the planning model (Fable) revises the plan → show it again. No iteration cap; user controls exit via `onayla` or `iptal`.
-   - Persist `clarificationRounds`, `clarificationQuestions`, `clarificationAnswers`, `planIterations`, `planApprovedAt`, `planEditRequests` to `state.phases["1"]`.
+Run all **8 sequential interactive steps** in `$HOME/.claude/multi-agent-refs/phases/phase-0-init.md`; never short-circuit. Step 0: `bash "$HOME/.copilot/lib/credential-inventory.sh" --json` into `agent-state.json.credentialInventory`; never ask for data a mapped credential can fetch. **Exit gate (BLOCKING)**: `node "$HOME/.copilot/scripts/phase0-exit-gate.mjs" "$TASK_ID" --input "$USER_INPUT"`; non-zero means Phase 0 skipped its own output (`taskType`, `baseBranch`, `baseFetchStatus`, a worktree distinct from the project root, the Figma tier for a Figma URL): fix the missing step. Per-project parsing and worktree: `$HOME/.claude/multi-agent-refs/orchestrator/phase-0-projects.md`. Jira input: `bash "$HOME/.copilot/lib/jira-family-write.sh" resolve --state "$STATE_FILE"` (records, writes nothing).
 
-### Phase 2: Dev (claude-sonnet-5-5)
-For each todo (respecting dependency order):
-1. Update todo status: `in_progress`
-2. **TDD cycle**: write failing test → verify RED; write minimal code → verify GREEN; refactor if needed
-3. Run build: `xcodebuild` via **task agent**
-4. If build fails → fix → rebuild (max 3 attempts, track `retryCount` in `agent-state.json`)
-5. Update todo status: `done`
-6. Log: `⚡ Phase 2: {todo-id}  -  Test ✅ Build ✅`
+### Phases 1-5
 
-### Phase 3: Review (parallel + triage)
-0. **Diff Risk Scoring (advisory)**  -  before reviewer dispatch run `node $HOME/.claude/scripts/diff-risk-score.mjs --base "$BASE_BRANCH" --top 5` and inject the top-N risk-ranked files as a `${PRIORITY_FILES}` block into each reviewer's prompt. Heuristic, deterministic, sub-second, never gates the pipeline. Disabled when `prefs.global.diffRiskAdvisory = false`. Signals: security paths (×3), schema migrations (×4), public API surfaces (×2), no-test-change (×2.5), complexity delta (×1.5), UI-critical paths (×1.5), loc changed (×1).
-1. Launch **code-reviewer** agents in parallel. Reviewer set depends on which CLI is hosting the pipeline:
-   - **Claude Code** (3 reviewers): `claude-fable-5-1` (deep security + architecture) + `claude-opus-5-5` (edge cases, different perspective) + `claude-sonnet-5-5` (quality + correctness)
-   - **Copilot CLI** (3 reviewers): `gpt-5.4` (edge cases, different perspective) + `claude-opus-4.7` + `claude-sonnet-4.6` (Fable is not offered on Copilot CLI)
-   - Triage: single top-tier pass over merged findings (`claude-fable-5-1` on Claude Code, `claude-opus-4.7` on Copilot CLI)
-2. Collect findings, classify:
-   - 🔴 **Blocking** → must fix → back to Phase 2 (max 3 iterations)
-   - 🟡 **Important** → fix and re-review
-   - 🟢 **Suggestion** → apply if reasonable
-3. Log: `🔍 Phase 3: Review  -  {consensus}`
+- **1 Plan**: analysis (`explorer`), planning, then the **Plan Approval Gate** (interactive only): clarification max 2 rounds, approval loop `onayla` / `iptal` / free-text.
+- **2 Dev**: TDD per todo, build, max 3 fix attempts.
+- **3 Review**: three parallel reviewers + one triage pass; blocking goes back to Phase 2 (max 3 iterations). Interactive user test via `AskUserQuestion`, saying yes removes the worktree and checks the branch out: `$HOME/.claude/multi-agent-refs/orchestrator/phase-3-user-test.md`.
+- **4 Commit & PR**: ask before commit (No pauses; `resume` later) and before the PR; `git add` specific files, never `git add -A`; author from `prefs.global.identities[]`. Read `$HOME/.claude/multi-agent-refs/orchestrator/phase-4-5-projects.md`.
+- **5 Report**: finalize `agent-log.md` (`$HOME/.claude/multi-agent-refs/phases/log-format.md`), print the summary.
 
-#### Phase 3 user test (optional, interactive)
-Lets the user test the code before the PR, through a native `AskUserQuestion` picker (never a typed y/N prompt). The prompt MUST say that yes **removes the worktree and checks the branch out locally**, or the side effect is implicit. Skip → Phase 4; a requested fix recreates the worktree and returns to Phase 2. Commits live on the branch, so removing the worktree does NOT delete them.
-
-Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/phase-3-user-test.md`. Read it when Phase 3 review is done and the run is interactive.
-
-### Phase 4: Commit & PR
-
-Every project:
-1. **Ask**: "Want to commit?" **No** → pause; the user continues with `resume` when ready.
-2. my-figma-app: run `/figma-iteration-commit {ComponentName}`  -  do NOT commit manually, the skill manages everything.
-3. my-app / generic: stage specific files with `git add` (NOT `git add -A`  -  avoid committing .env, credentials, or other sensitive files), commit as `{type}({scope}): {description} [PROJ-{id}]`, push, then **Ask**: "Want to open a Pull Request?" before creating it.
-4. Author: resolved from `prefs.global.identities[]`  -  NEVER hardcode.
-
-Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/phase-4-5-projects.md`. Read it when Phase 4 starts, for the per-project steps and the GitHub issue body update.
-
-### Phase 5: Report
-1. Finalize `agent-log.md` with ALL sections:
-   - **Timeline table**: phase, agent, model, duration, status, detail
-   - **Agent Activity Report**: per-agent call count, total duration, models used, skill count
-   - **Token Estimate**: estimated token usage per agent + total
-   - **Per-Phase Time Distribution**: ASCII bar chart showing time per phase
-   - **Review Iterations**: blocking/important/suggestion counts per iteration
-   - **Files Changed**: `git diff --stat` output
-   - **Review Consensus**: final 🔴/🟡/🟢 status
-   - **Test Scenarios**: Turkish, for Jira (Precondition → Steps → Expected result)
-2. Print compact summary to terminal:
-   ```
-   ✅ PROJ-{id} complete
-   ⏱️  Total: 3m 18s | Agents: 12 calls | Files: 4 changed
-   📦 Commit: abc1234 | PR: #87
-   📖 Wiki: {ComponentName}.md pushed to wiki master
-   📋 Full report: ~/.claude/logs/multi-agent/{project}/{task-id}/agent-log.md
-   ```
-3. Log: `📋 Phase 5: Report complete`
-
-**Step 2  -  Wiki + Figma screenshots** runs only for a component implementation with a GitHub Issue (skip for a bug fix or refactor). Full procedure: `$HOME/.claude/multi-agent-refs/orchestrator/phase-4-5-projects.md`. Read it when Phase 5 reports a component task.
-
----
-
-## Log File Format
-
-Canonical shape, and the path the code actually uses, live in
-`$HOME/.claude/multi-agent-refs/phases/log-format.md`. Read it when writing
-the log; do not re-derive the format here.
-
-## Component Generation
-
-Generic guide: `$HOME/.claude/multi-agent-refs/component-generation.md`.
-Dispatch (which plugin skill handles it per CLI):
-`$HOME/.claude/multi-agent-refs/component-dispatch.md`.
-Loaded only for a task that generates a component from a design.
+Component generation: `$HOME/.claude/multi-agent-refs/component-generation.md` and `component-dispatch.md`, only for a design-to-component task.
 
 ## Rules
 
-- ❌ NEVER put "Copilot", "AI", "generated by" in code or commits
-- ❌ NEVER commit without passing build
-- ❌ NEVER commit without passing review
-- ❌ NEVER auto-close an issue (GitHub or Jira); PR bodies use `Ref: #N` / `Related: #N`, never `Closes/Fixes/Resolves`
-- ✅ Follow existing code style and conventions
-- ✅ Use design tokens, no magic numbers
-- ✅ Every public method must have tests
-- ✅ Xcode header on new Swift files
-- ✅ Commit message format:
-  - With Jira: `{type}({scope}): description [PROJ-{id}]`
-  - Without Jira (generic): `{type}({scope}): description [#{shortId}]`
-- ✅ Git author: resolved from `prefs.global.identities[]` (e.g. `{identity.name} <{identity.email}>`)  -  NEVER hardcoded
-
-## Help Display
+- NEVER put "Copilot", "AI", "generated by" in code or commits
+- NEVER commit without passing build or review
+- NEVER auto-close an issue (GitHub or Jira); PR bodies use `Ref: #N` / `Related: #N`, never `Closes/Fixes/Resolves`
+- Follow existing code style; design tokens, no magic numbers; every public method tested; Xcode header on new Swift files
+- Commit: `{type}({scope}): description [PROJ-{id}]`, or `[#{shortId}]` without Jira
+- Git author from `prefs.global.identities[]`, never hardcoded
 
-Rendered by the `multi-agent-help` skill (`/multi-agent-help`), which owns the full text in both languages.
+Help: the `multi-agent-help` skill (`/multi-agent-help`).