maestro-flow 0.5.36 → 0.5.37

package/.codex/skills/maestro/SKILL.md

@@ -1,565 +1,565 @@
----
-name: maestro
-description: Auto-route intent to optimal command chain
-argument-hint: "\"intent text\" [-y] [-c|--continue] [--dry-run] [--super]"
-allowed-tools: Read, Write, Edit, Bash, Glob, Grep, request_user_input
----
-
-<purpose>
-Sequential pipeline coordinator. Classify intent → decompose (broad lifecycle intents) →
-resolve chain → **directly invoke each skill in-context, one at a time** → report.
-
-Entry points:
-- **`$maestro "intent"`** — Classify → decompose → chain → execute
-- **`$maestro --continue`** — Resume from first pending step
-- **`$maestro --dry-run "intent"`** — Show chain, no execution
-- **`$maestro --super "intent"`** — Production-ready mode (read maestro-super.md)
-
-Codex specifics (parity with maestro-ralph):
-- **No agent spawning** — skills run directly in coordinator context, sequentially.
-- **Goal created via built-in tool** — `create_goal` binds the decomposed sub-goal checklist;
-  `update_plan` mirrors steps; `update_goal` releases on convergence.
-</purpose>
-
-<deferred_reading>
-- [maestro-super.md](~/.maestro/workflows/maestro-super.md) — read when `--super` flag is active
-</deferred_reading>
-
-<context>
-$ARGUMENTS — user intent text, or special flags.
-
-**Flags:**
-- `-y, --yes` — Auto mode: skip all prompts; propagate `-y` to each skill
-- `--continue` — Resume latest paused session from first pending step
-- `--dry-run` — Display planned chain without executing
-- `--super` — Read and follow `maestro-super.md` completely
-
-**Session state**: `.workflow/.maestro/{session-id}/`
-</context>
-
-<invariants>
-1. **Skills invoked DIRECTLY in-context** — coordinator runs `$skill {resolved_args}` itself, sequentially. NO spawn_agents_on_csv, NO wave/CSV/worker.
-2. **Coordinator owns the loop** — classify → decompose → resolve chain → for each step: resolve args → invoke skill → read result → persist → next.
-3. **Decomposition contract shared with maestro-ralph** — broad/lifecycle intents run S_DECOMPOSE producing the SAME additive block (`boundary_contract`, `execution_criteria`, `task_decomposition`). Reference maestro-ralph `A_DECOMPOSE_TASKS`
-4. **Goal is tool-created** — `A_DECOMPOSE_TASKS` calls `create_goal` with sub-goal success criteria. `update_goal` on convergence; held while aborted/paused
-5. **status.json 唯一真源** — 不生成 `goal-checklist.md`；step 含 `command_scope` + `command_path` + `completion_confirmed`
-6. **Topology awareness** — chain catalog 含 grill / brainstorm / blueprint / analyze-macro(text) / analyze(numeric) / roadmap / plan(三路径) / execute / ...
-6.5. **Grill is interactive-only** — auto_mode MUST skip grill stage and route directly to brainstorm; grill requires Socratic Q&A with the user
-7. **D-007 milestone 反查** — 数字 phase 步骤的 `milestone_id` 由 `state.json.milestones[].phase_slugs` 反查
-8. **schema 向后兼容** — decomposition 字段可选；`steps[]` 由 post-goal-audit 动态生长（goal_ref tagged）；既有字段不删不改；`waves` 保留空数组
-9. **Sequential execution** — one step at a time in index order; each step's result read before the next starts
-10. **Abort on failure** — failed step → mark remaining skipped → report (goal stays bound for `--continue`)
-</invariants>
-
-<state_machine>
-
-<states>
-S_PARSE         — 解析参数、检测 flags              PERSIST: —
-S_CONTINUE      — 加载已有 session，定位 resume 点   PERSIST: session (loaded)
-S_CLASSIFY      — 意图分类、解析 chain (A_CLASSIFY)   PERSIST: —
-S_DECOMPOSE     — 边界澄清、写执行准则+子目标、建 goal PERSIST: session.boundary_contract, .execution_criteria, .task_decomposition
-S_CREATE        — 创建 session + status.json         PERSIST: session.status, session.steps[]
-S_DRY_RUN       — 显示 chain 后结束                  PERSIST: —
-S_CONFIRM       — 用户确认（auto_mode 跳过）          PERSIST: —
-S_STEP_LOOP     — 逐步直接调用 skill → 读结果 → 循环  PERSIST: session.current_step, session.steps[], session.context
-S_DECISION_EVAL — 评估 post-goal-audit 决策节点       PERSIST: —
-S_COMPLETE      — 标记完成、释放目标                  PERSIST: session.status = "completed"
-S_ABORTED       — 失败中止、标记剩余 skipped          PERSIST: session.status = "aborted"
-S_FALLBACK      — 意图无法分类，请求输入              PERSIST: —
-</states>
-
-<transitions>
-
-S_PARSE:
-  → S_CONTINUE    WHEN: --continue flag
-  → S_CLASSIFY    WHEN: intent text present
-  → S_FALLBACK    WHEN: no intent AND no flags
-
-S_CONTINUE:
-  → S_STEP_LOOP   WHEN: session found, has pending steps     DO: A_RESUME_SESSION
-  → S_FALLBACK    WHEN: no session found
-
-S_CLASSIFY:
-  → S_DECOMPOSE   WHEN: chain resolved                      DO: A_CLASSIFY
-  → S_FALLBACK    WHEN: no match AND auto_mode
-  → S_CLASSIFY    WHEN: no match AND not auto_mode          DO: A_CLARIFY_INTENT
-                   GUARD: max 1 clarification attempt → S_FALLBACK
-
-S_DECOMPOSE:
-  → S_CREATE      DO: A_DECOMPOSE_TASKS
-                   GUARD: broad intent (重构/全面/重写/迁移/overhaul/migrate/rewrite) on multi-step lifecycle chain → MUST clarify even if auto_mode
-                   GUARD: single-step chain OR narrow intent OR chain ∈ {status,init,quick} → skip decomposition (pass through)
-
-S_CREATE:
-  → S_DRY_RUN     WHEN: --dry-run flag                      DO: A_CREATE_SESSION
-  → S_CONFIRM     WHEN: not auto_mode                       DO: A_CREATE_SESSION
-  → S_STEP_LOOP   WHEN: auto_mode                           DO: A_CREATE_SESSION
-
-S_DRY_RUN:
-  → END           DO: display chain with step types + sub-goal summary
-
-S_CONFIRM:
-  → S_STEP_LOOP   WHEN: user confirms
-  → S_ABORTED     WHEN: user cancels
-
-S_STEP_LOOP:
-  → S_DECISION_EVAL WHEN: next step.type == "decision"
-  → S_STEP_LOOP   WHEN: next step.type == "skill"           DO: A_EXEC_STEP
-  → S_COMPLETE    WHEN: no pending steps
-  → S_ABORTED     WHEN: step failed (auto_mode: retry once then abort)
-
-S_DECISION_EVAL:                                            ENTRY: A_GOAL_AUDIT_EVALUATE (produces verdict)
-  → S_STEP_LOOP   WHEN: verdict == all_met                  DO: A_APPLY_GOAL_DONE
-  → S_STEP_LOOP   WHEN: verdict == has_unmet                DO: A_APPLY_GOAL_FIX
-  → S_ABORTED     WHEN: retry >= max_retries AND unmet      DO: escalate (insert quality-debug "{gaps}")
-
-S_COMPLETE:
-  → END           DO: A_FINALIZE
-
-S_ABORTED:
-  → END           DO: A_ABORT_REPORT
-
-S_FALLBACK:
-  → S_CLASSIFY    WHEN: user provides new intent            DO: request_user_input
-  → END           WHEN: user cancels
-
-</transitions>
-
-<actions>
-
-### A_CLASSIFY
-
-**Layer 1: Exact-match (fast path)**
-- `--chain <name>` flag → validate against chainMap, use directly (E002 if not found)
-- `continue`/`next`/`go`/`继续`/`下一步` → `state_continue`
-- `status`/`状态`/`dashboard` → `status`
-
-If matched, skip to chain resolution.
-
-**Layer 2: Semantic intent matching**
-
-Directly match user intent to the best `task_type` (maps to chain in Chain Map). Use LLM semantic understanding — no rigid keyword lookup.
-
-Extract:
-```json
-{
-  "task_type": "<from chain catalog below>",
-  "scope":     "<module/file/area or null>",
-  "issue_id":  "<ISS-XXXXXXXX-NNN if mentioned, else null>",
-  "phase_ref": "<integer if mentioned, else null>",
-  "urgency":   "<low|normal|high>"
-}
-```
-
-**Chain catalog — select by best semantic fit:**
-
-| task_type | When user intent is about... |
-|-----------|---------------------------|
-| `grill` | Stress-test, challenge assumptions, Socratic questioning on a plan/idea (**skip when auto_mode — grill is interactive-only**) |
-| `quick` | Simple/small task, add a feature, quick change |
-| `blueprint` | Formal spec generation (Product Brief / PRD / Architecture / Epics) |
-| `analyze_macro` | Broad/medium intent w/o numeric phase — explore impact, produce scope_verdict |
-| `plan_from_analyze` | Plan directly from analyze artifact (no roadmap, scope=standalone) |
-| `plan_from_blueprint` | Plan directly from blueprint artifact (scope=standalone) |
-| `plan` | Plan, design, architect a phase |
-| `execute` | Implement, develop, code a phase |
-| `analyze` | Understand, investigate, evaluate code (numeric phase) |
-| `verify` | Check goals met, validate results (routes to quality-review) |
-| `review` | Code quality review |
-| `test` | Run or create tests, UAT |
-| `test_gen` | Generate tests for coverage gaps |
-| `debug` | Diagnose, troubleshoot, fix broken behavior |
-| `refactor` | Restructure, clean up, reduce tech debt |
-| `init` | Initialize project |
-| `sync` | Update/sync documentation |
-| `retrospective` | Phase review, post-mortem, 复盘 |
-| `learn` | Capture insights, record learnings |
-| `release` | Publish, ship, tag version |
-| `amend` | Revise workflow commands |
-| `compose` | Design/compose reusable workflows |
-| `overlay` | Create/edit command overlays |
-| `update` | Update maestro itself |
-| `harvest` | Extract knowledge from artifacts |
-| `domain_add` | Register a domain term into glossary |
-| `domain_list` | List registered domain terms |
-| `domain_discover` | Discover domain term candidates from codebase |
-| `wiki` | Manage wiki graph |
-| `knowhow` | Manage knowhow entries |
-| `ui_design` | UI design, build new UI |
-| `issue` | Issue CRUD — create, list, close, query |
-| `issue_discover` | Discover/find issues in codebase |
-| `issue_analyze` | Analyze a specific issue |
-| `issue_plan` | Plan fix for an issue |
-| `issue_execute` | Fix issue end-to-end (auto-upgrades to issue-full) |
-| `feature` | Standard feature: plan→execute→review |
-| `full-lifecycle` | Complete phase: plan→execute→review→test→audit→complete |
-| `brainstorm-driven` | Start from exploration/brainstorm |
-| `spec-driven` | From spec/requirements (heavy, with init) |
-| `roadmap-driven` | From requirements (light, with init) |
-| `analyze-plan-execute` | Fast track: analyze→plan→execute |
-| `execute-review` | Resume after planning |
-| `review-fix` | Fix review-blocked issues |
-| `quality-loop` | Full quality improvement cycle |
-| `quality-loop-partial` | Partial quality fix |
-| `quality-fix` | Analyze gaps→plan→execute→review |
-| `deploy` | Verify then release |
-| `milestone-close` | Close/transition milestone |
-| `milestone-release` | Release milestone with version tag |
-| `phase_transition` | Transition phase: audit→complete |
-| `next-milestone` | Advance to next milestone |
-| `state_continue` | Continue from current project state |
-
-**Selection priorities:**
-1. `issue_id` present → prefer issue chains
-2. UI/design/界面/页面/原型 → prefer `ui_design`
-3. 正式规格/spec-generate/7-phase → `blueprint` (single-step) 或 `blueprint-driven`
-4. 压力测试/拷问/grill/stress-test → `grill` (single-step); **auto_mode → skip grill, route to `brainstorm-driven` instead**
-5. 头脑风暴/探索 → `brainstorm-driven`
-5. Broad/medium intent + 无数字 phase → `analyze_macro`（产 scope_verdict）；后续 large→roadmap链；medium/small→`plan_from_analyze`
-6. 已有 analyze artifact 直达 plan → `plan_from_analyze`
-7. 已有 blueprint artifact 直达 plan → `plan_from_blueprint`
-8. Multiple lifecycle steps implied → prefer multi-step chains
-9. Single specific action → prefer single-step chains
-10. "问题" describing broken behavior → `debug`; tracked item with ISS-ID → `issue`
-11. Simple task, no lifecycle context → `quick`
-12. Global fallback → `quick`
-
-**Clarity scoring**: 3=task_type+scope+phase, 2=task_type+scope, 1=task_type only, 0=empty.
-If `clarity < 2` and not `auto_mode` → transition to A_CLARIFY_INTENT.
-
-**Layer 4: State-based routing** (when `taskType === 'state_continue'`)
-
-Read `.workflow/state.json` and route by condition:
-
-| Condition | Chain |
-|-----------|-------|
-| Not initialized | `init` |
-| No phases, no roadmap, has accumulated_context | `next-milestone` |
-| No phases | `brainstorm-driven` |
-| pending + has context | `plan` |
-| pending, no context | `analyze` |
-| exploring/planning + has plan | `execute-review` |
-| exploring/planning, no plan | `plan` |
-| executing, all tasks done | `review` |
-| executing, tasks remain | `execute` |
-| reviewing, verdict == BLOCK | `review-fix` |
-| reviewing, verdict != BLOCK + UAT pending | `test` |
-| reviewing, verdict != BLOCK + UAT passed | `milestone-close` |
-| reviewing, verdict != BLOCK + UAT failed | `debug` |
-| testing, UAT passed | `milestone-close` |
-| testing, UAT not passed | `debug` |
-| completed | `milestone-close` |
-| blocked | `debug` |
-| fallback | `status` |
-
-**Chain resolution order:**
-1. `forceChain` → `chainMap[forceChain]` (E002 if not found)
-2. `state_continue` → Layer 4 state routing → `{ chain, argsOverride? }`
-3. `taskToChain[taskType]` → alias lookup (see Chain Aliases below)
-4. `chainMap[taskType]` → direct lookup
-
-**Phase resolution**: structured extraction `phase_ref` → fallback regex (`phase N` or bare number) → `projectState.current_phase`.
-
-### A_CLARIFY_INTENT
-
-1. `request_user_input` with available chain types
-2. Re-classify with user response
-
-### A_DECOMPOSE_TASKS
-
-与 maestro-ralph `A_DECOMPOSE_TASKS` 共享分解契约。Condensed:
-
-1. 分类意图广度。narrow / 单步 / `{status,init,quick}` 链跳过
-2. broad/medium → `request_user_input` ≤3 轮：Scope / Constraints / Definition of Done
-3. 派生 `execution_criteria` + `task_decomposition`（每个 sub-goal 含 `done_when` + `evidence` + `lifecycle` + `completion_confirmed: false`）
-4. **status.json 唯一真源**：写入 `boundary_contract` / `execution_criteria` / `task_decomposition`；不生成 markdown 清单
-5. 链路末尾（evidence 产出步骤后、milestone-complete/close-out 前）追加 `decision:post-goal-audit`。S_DECISION_EVAL 据此动态生长 `steps[]`
-6. **Register goal via `create_goal`:**
-   ```
-   create_goal({ objective: "Maestro {chain}: {intent} — converge {N} sub-goals within boundary",
-     success_criteria: task_decomposition.map(g => `${g.id}: ${g.done_when}`),
-     constraints: [...execution_criteria, "stay within boundary_contract; resume via $maestro --continue"] })
-   ```
-
-### A_CREATE_SESSION
-
-1. Read `.workflow/state.json` 获取 phase / milestone（D-007 反查 `phase_slugs`）；读最新 macro analyze artifact 注入 `scope_verdict` + `analyze_macro_id`；读最新 blueprint artifact 注入 `blueprint_id`
-2. Resolve chain's skill list from Chain Map (see appendix)
-3. **Prevalidate via `Bash("maestro ralph skills --platform codex --json --quiet")`** 一次性拉取所有可用 codex skills（global `~/.codex/skills/` + project `.codex/skills/`，project 覆盖 global），匹配 skill 名得到：
-   - 命中 → `command_scope = "global" | "project"`，`command_path = <绝对 SKILL.md 路径>`
-   - 未命中 → `command_scope = "missing"`, `command_path = null`
-4. Create `.workflow/.maestro/maestro-{YYYYMMDD-HHMMSS}/status.json`（与 ralph 共用 schema）:
-   ```json
-   {
-     "session_id", "source": "maestro", "intent", "task_type", "chain_name",
-     "phase", "phase_is_new": false, "milestone": "",
-     "scope_verdict": null, "analyze_macro_id": null, "blueprint_id": null,
-     "auto_mode": false,
-     "context": { "issue_id": null, "scratch_dir": null, "plan_dir": null,
-       "analysis_dir": null, "brainstorm_dir": null, "blueprint_dir": null, "grill_id": null },
-     "steps": [{
-       "index": 0, "type": "skill|decision",
-       "skill": "", "args": "",
-       "stage": "", "scope": null,
-       "command_scope": "global|project|missing|null",
-       "command_path": "~/.codex/skills/{name}/SKILL.md | .codex/skills/{name}/SKILL.md | null",
-       "milestone_id": null, "source_artifact_ref": null,
-       "status": "pending", "goal_ref": null,
-       "completion_confirmed": false, "completion_status": null,
-       "completion_evidence": null, "completed_at": null
-     }],
-     "waves": [], "current_step": 0, "status": "running",
-     "boundary_contract": {}, "execution_criteria": [],
-     "task_decomposition": [{ "id": "G1", "goal": "", "done_when": "", "evidence": "",
-       "status": "pending|done", "completion_confirmed": false, "completed_at": null }],
-     "task_decomposition_all_done": false
-   }
-   ```
-   Decomposition fields written ONLY if A_DECOMPOSE_TASKS produced them (additive)
-5. Validate: 所有 step 的 `command_scope != "missing"`；否则 raise E006 列出缺失 skill
-6. Initialize tracking:
-   - If decomposed: goal already registered by A_DECOMPOSE_TASKS. Else: `create_goal({ objective: "Maestro {chain}: {N} steps [{skill list}]" })`
-   - `update_plan({ plan: steps.map(step => ({ step, status: "pending" })) })`
-
-### A_RESUME_SESSION
-
-1. Glob `.workflow/.maestro/maestro-*/status.json` sorted desc, load most recent
-2. Find first pending step → set as resume point
-3. Rebuild `update_plan` from status.json (completed→"completed", current→"in_progress", rest→"open")
-
-### A_EXEC_STEP
-
-Direct in-context skill invocation — **replaces the old spawn/wave/CSV mechanism**.
-
-1. **buildSkillCall**: replace placeholders + append auto-yes flag if `auto_mode` (see Appendix: Auto-Yes Flag Map):
-
-   | Placeholder | Source |
-   |-------------|--------|
-   | `{phase}` | session.phase |
-   | `{plan_dir}` | session.context.plan_dir |
-   | `{analysis_dir}` | session.context.analysis_dir |
-   | `{brainstorm_dir}` | session.context.brainstorm_dir |
-   | `{spec_session_id}` | session.context.spec_session_id |
-   | `{GRL}` | session.context.grill_id |
-   | `{ANL}` | session.analyze_macro_id |
-   | `{BLP}` | session.blueprint_id |
-   | `{intent}` | session.intent |
-
-   **--from auto-injection**: 当 step 是 `maestro-plan`，args 含 `{phase}` 但无 `--from` 且无 `--dir`，且 `session.context.analysis_dir` 已填充 → 查 state.json 同 phase+milestone 最新 completed analyze artifact → 注入 `--from analyze:{id}`，写 `step.source_artifact_ref`
-2. Mark step `status="running"`, persist status.json + `update_plan` (this step → in_progress)
-3. **Invoke the skill directly**: execute `$skill {resolved_args}` in coordinator context (NO spawn). Read its produced artifacts directly
-4. On success: capture summary; mark step `status="done"`. **Barrier-context update** (when step is a context-producing skill):
-   | Skill | Read | Context Updates |
-   |-------|------|-----------------|
-   | maestro-grill | grill-report.md, state.json | grill_id |
-   | maestro-analyze | context.md, state.json | analysis_dir, gaps, phase |
-   | maestro-plan | plan.json, .task/TASK-*.json | plan_dir, task_count |
-   | maestro-brainstorm | .brainstorming/ | brainstorm_dir, features |
-   | maestro-roadmap | specs/ | spec_session_id |
-   | maestro-execute | results.csv | exec_completed, exec_failed |
-5. On failure: mark `status="failed"`; auto_mode → retry once → still failed → S_ABORTED
-6. Persist status.json + `update_plan` after every step
-
-### A_GOAL_AUDIT_EVALUATE
-
-S_DECISION_EVAL 入口；镜像 maestro-ralph `A_GOAL_AUDIT_EVALUATE`。Condensed:
-
-1. 读 `session.task_decomposition`（status.json，真源）
-2. 对每个 `status != "done"` 的子目标：解析 `evidence` 产物
-3. `maestro delegate --role analyze --mode analysis` 读取 evidence、对照 done_when 判定，返回 `STATUS=all_met|has_unmet / UNMET=[{id,gap,target_phase}] / CONFIDENCE_SCORE`
-4. status.json 为写入目标：每个达成子目标 `status="done"` + `completed_at=now`；然后从 status.json 重渲染 checklist（Sync Rule）
-5. Verdict（`all_met` / `has_unmet`）由 S_DECISION_EVAL 消费。GUARD: retry >= max_retries AND still unmet → escalate
-
-### A_APPLY_GOAL_FIX
-
-**Dynamic step-growth core** (mirrors maestro-ralph). For each unmet sub-goal (grouped by target_phase), insert before the post-goal-audit node a scoped mini-loop `$maestro-plan --gaps {phase} "G{n}: {gap}" → $maestro-execute {phase}`, each tagged `goal_ref: "G{n}"`, type `"skill"`. Re-append `decision:post-goal-audit {retry+1}`. Reindex, increment retry, persist + `update_plan`. `steps[]` grew.
-
-### A_APPLY_GOAL_DONE
-
-1. status.json：全部 `task_decomposition[*].status="done"` + `completion_confirmed=true` + `completed_at=now` + `task_decomposition_all_done=true`
-2. `update_goal({ status: "complete" })`
-3. 继续到 chain 的终结步骤
-
-### A_FINALIZE
-
-1. Set `session.status = "completed"`, write status.json
-2. Sync `update_plan`: all steps → "completed"
-3. `update_goal({ status: "complete" })` — release goal (idempotent if already released)
-4. Generate completion report (see Appendix: Report Format)
-
-### A_ABORT_REPORT
-
-1. Mark remaining steps `skipped` in status.json
-2. Set `session.status = "aborted"`, write status.json; sync `update_plan`
-3. Do NOT call `update_goal` — goal stays for `--continue` resume
-4. Display abort report with failure details
-
-</actions>
-
-</state_machine>
-
-<appendix>
-
-### Chain Map (Full)
-
-**Single-step chains:**
-
-| Chain | Command + Args |
-|-------|---------------|
-| `grill` | `maestro-grill "{intent}"` |
-| `status` | `manage-status` |
-| `init` | `maestro-init` |
-| `blueprint` | `maestro-blueprint "{intent}"` |
-| `analyze_macro` | `maestro-analyze "{intent}"` |
-| `analyze` | `maestro-analyze {phase}` |
-| `ui_design` | `maestro-impeccable build "{phase}"` |
-| `plan` | `maestro-plan {phase}` |
-| `plan_from_analyze` | `maestro-plan --from analyze:{analyze_macro_id}` |
-| `plan_from_blueprint` | `maestro-plan --from blueprint:{blueprint_id}` |
-| `execute` | `maestro-execute {phase}` |
-| `verify` | `quality-review {phase}` |
-| `test_gen` | `quality-auto-test {phase}` |
-| `auto_test` | `quality-auto-test {phase}` |
-| `test` | `quality-test {phase}` |
-| `debug` | `quality-debug "{description}"` |
-| `integration_test` | `quality-auto-test {phase}` |
-| `refactor` | `quality-refactor "{description}"` |
-| `review` | `quality-review {phase}` |
-| `retrospective` | `quality-retrospective {phase}` |
-| `learn` | `maestro-learn "{description}"` |
-| `sync` | `quality-sync` |
-| `milestone_audit` | `maestro-milestone-audit` |
-| `milestone_complete` | `maestro-milestone-complete` |
-| `codebase_rebuild` | `manage-codebase-rebuild` |
-| `codebase_refresh` | `manage-codebase-refresh` |
-| `spec_setup` | `spec-setup` |
-| `spec_add` | `spec-add "{description}"` |
-| `spec_load` | `spec-load` |
-| `spec_map` | `manage-codebase-rebuild` |
-| `spec_remove` | `spec-remove "{description}"` |
-| `domain_add` | `domain-add "{description}"` |
-| `domain_list` | `domain-list` |
-| `domain_discover` | `domain-discover` |
-| `knowhow_capture` | `manage-knowhow-capture "{description}"` |
-| `knowhow` | `manage-knowhow "{description}"` |
-| `issue` | `manage-issue "{description}"` |
-| `issue_discover` | `manage-issue-discover "{description}"` |
-| `issue_analyze` | `maestro-analyze --gaps "{description}"` |
-| `issue_plan` | `maestro-plan --gaps` |
-| `issue_execute` | `maestro-execute` |
-| `quick` | `maestro-quick "{description}"` |
-| `harvest` | `manage-harvest "{description}"` |
-| `wiki` | `manage-wiki` |
-| `wiki_connect` | `wiki-connect` |
-| `wiki_digest` | `wiki-digest` |
-| `business_test` | `quality-auto-test {phase}` |
-| `amend` | `maestro-amend "{description}"` |
-| `release` | `maestro-milestone-release` |
-| `compose` | `maestro-composer "{description}"` |
-| `play` | `maestro-player "{description}"` |
-| `update` | `maestro-update` |
-| `overlay` | `maestro-overlay "{description}"` |
-| `link_coordinate` | `maestro-link-coordinate "{description}"` |
-
-**Multi-step chains:**
-
-| Chain | Steps (→ = sequential, [B] = context-producing barrier) |
-|-------|---------------------------------------|
-| `feature` | [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `quality-fix` | [B] maestro-analyze --gaps → [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
-| `deploy` | quality-review → maestro-milestone-release |
-| `blueprint-driven` | maestro-init → [B] maestro-blueprint → [B] maestro-plan --from blueprint:{BLP} → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `analyze-macro-driven` | [B] maestro-analyze "{intent}" → ◆ post-analyze-scope → (large: [B] maestro-roadmap --from analyze:{ANL} → [B] maestro-analyze {phase} → [B] maestro-plan {phase}) / (medium\|small: [B] maestro-plan --from analyze:{ANL}) → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `grill-brainstorm` | [B] maestro-grill → [B] maestro-brainstorm --from grill:{GRL} → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto (**auto_mode: skip grill step, fall back to brainstorm-driven**) |
-| `brainstorm-driven` | [B] maestro-brainstorm → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `ui-craft-build` | maestro-impeccable build → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `roadmap-driven` | maestro-init → [B] maestro-roadmap → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `next-milestone` | [B] maestro-roadmap → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
-| `full-lifecycle` | [B] maestro-plan → [B] maestro-execute → quality-review → quality-test → maestro-milestone-audit → maestro-milestone-complete → manage-harvest --auto |
-| `execute-review` | [B] maestro-execute → quality-review |
-| `analyze-plan-execute` | [B] maestro-analyze -q → [B] maestro-plan --dir {scratch_dir} → [B] maestro-execute --dir {scratch_dir} → manage-harvest --auto |
-| `quality-loop` | quality-review → quality-test → quality-debug --from-uat → [B] maestro-plan --gaps → [B] maestro-execute |
-| `quality-loop-partial` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
-| `review-fix` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
-| `milestone-close` | maestro-milestone-audit → maestro-milestone-complete |
-| `milestone-release` | maestro-milestone-audit → maestro-milestone-release |
-| `phase_transition` | maestro-milestone-audit → maestro-milestone-complete |
-| `issue-full` | [B] maestro-analyze --gaps → [B] maestro-plan --gaps → [B] maestro-execute → quality-review → manage-issue close → manage-harvest --auto |
-| `issue-quick` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review → manage-issue close |
-
-> When S_DECOMPOSE ran, a `decision:post-goal-audit` node is appended as the final node (after the last evidence-producing step; before milestone-complete/close-out if the chain ends with one). `[B]` now denotes a context-producing skill (artifacts read into `session.context`) — execution is still sequential (no parallelism; spawning removed).
-
-**Chain Aliases** (taskType → chain):
-
-| taskType | Chain |
-|----------|-------|
-| `spec_generate` | `blueprint-driven` |
-| `spec-driven` | `blueprint-driven` |
-| `brainstorm` | `brainstorm-driven` |
-| `issue_execute` | `issue-full` |
-| `analyze_macro` | `analyze-macro-driven` |
-
-### Auto-Yes Flag Map
-
-| Skill | Flag |
-|-------|------|
-| maestro-init, maestro-analyze, maestro-brainstorm, maestro-blueprint, maestro-impeccable, maestro-roadmap | `-y` |
-| maestro-plan, maestro-execute, maestro-milestone-complete | `-y` |
-| quality-auto-test, quality-retrospective | `-y` |
-| quality-test | `-y --auto-fix` |
-
-### Context-Producing Skills
-
-`maestro-analyze`, `maestro-plan`, `maestro-brainstorm`, `maestro-roadmap`, `maestro-execute` — their artifacts are read into `session.context` after the step completes (see A_EXEC_STEP step 4). Other skills produce no coordinator context. No parallelism — all steps run sequentially.
-
-### Error Codes
-
-| Code | Severity | Condition | Recovery |
-|------|----------|-----------|----------|
-| E001 | error | Intent unclassifiable after clarification | Default to `feature` chain |
-| E002 | error | Intent unresolvable after retry | List chains, abort |
-| E003 | error | Step skill invocation failed | auto_mode retry once, then abort chain |
-| E004 | error | Context artifact not found | Retry step once, then abort |
-| E005 | error | --continue: no session found | List sessions, prompt |
-| E006 | error | command_scope == "missing" for one or more steps | List missing skills, abort build |
-| W001 | warning | Context artifact partial | Continue with available context |
-
-### Success Criteria
-
-- [ ] Intent classified and chain resolved
-- [ ] Chain catalog 覆盖 blueprint / analyze_macro / plan_from_analyze / plan_from_blueprint / blueprint-driven / analyze-macro-driven 等新拓扑路径
-- [ ] D-007: 数字 phase 步骤的 `milestone_id` 通过 `state.json.milestones[].phase_slugs` 反查；写入 step
-- [ ] plan step args 支持 `{phase}` / `--from analyze:{ANL_ID}` / `--from blueprint:{BLP_ID}` 三路径，`source_artifact_ref` 写入
-- [ ] Broad lifecycle intents decomposed (≤3 boundary questions); narrow/single-step skip
-- [ ] Goal registered via built-in `create_goal`; status.json decomposition fields additive-only
-- [ ] status.json 唯一真源；无 markdown 清单
-- [ ] 每个 step 含 `command_scope` + `command_path` + `completion_confirmed` 字段
-- [ ] post-goal-audit node appended as final node; unmet sub-goals dynamically grow steps[] (goal_ref tagged)
-- [ ] Session dir initialized with status.json before first step
-- [ ] Every skill invoked DIRECTLY in-context — NO spawn_agents_on_csv, NO wave/CSV/worker
-- [ ] Sequential execution; status.json + update_plan persisted after every step
-- [ ] Context-producing skills' artifacts read into session.context before next step's args assembled
-- [ ] Failed step → remaining skipped → abort reported (goal held for --continue)
-- [ ] --dry-run shows chain + sub-goal summary, no execution
-- [ ] --continue resumes from first pending step
-- [ ] update_goal released on convergence (A_APPLY_GOAL_DONE / A_FINALIZE); held while aborted
-
-### Report Format
-
-```
-=== MAESTRO COMPLETE ===
-Session:  {sessionId}
-Chain:    {chain}
-Steps:    {completed}/{total}   Sub-goals: {done}/{total}
-
-STEP RESULTS:
-  [1] $maestro-analyze --gaps  →  ✓  found 3 gaps
-  [2] $maestro-plan --gaps     →  ✓  12 tasks
-  [◆] post-goal-audit          →  ✓  all sub-goals met
-  ...
-
-State:    .workflow/.maestro/{sessionId}/status.json
-Resume:   $maestro --continue
-```
-
-</appendix>
+---
+name: maestro
+description: Auto-route intent to optimal command chain
+argument-hint: "\"intent text\" [-y] [-c|--continue] [--dry-run] [--super]"
+allowed-tools: Read, Write, Edit, Bash, Glob, Grep, request_user_input
+---
+
+<purpose>
+Sequential pipeline coordinator. Classify intent → decompose (broad lifecycle intents) →
+resolve chain → **directly invoke each skill in-context, one at a time** → report.
+
+Entry points:
+- **`$maestro "intent"`** — Classify → decompose → chain → execute
+- **`$maestro --continue`** — Resume from first pending step
+- **`$maestro --dry-run "intent"`** — Show chain, no execution
+- **`$maestro --super "intent"`** — Production-ready mode (read maestro-super.md)
+
+Codex specifics (parity with maestro-ralph):
+- **No agent spawning** — skills run directly in coordinator context, sequentially.
+- **Goal created via built-in tool** — `create_goal` binds the decomposed sub-goal checklist;
+  `update_plan` mirrors steps; `update_goal` releases on convergence.
+</purpose>
+
+<deferred_reading>
+- [maestro-super.md](~/.maestro/workflows/maestro-super.md) — read when `--super` flag is active
+</deferred_reading>
+
+<context>
+$ARGUMENTS — user intent text, or special flags.
+
+**Flags:**
+- `-y, --yes` — Auto mode: skip all prompts; propagate `-y` to each skill
+- `--continue` — Resume latest paused session from first pending step
+- `--dry-run` — Display planned chain without executing
+- `--super` — Read and follow `maestro-super.md` completely
+
+**Session state**: `.workflow/.maestro/{session-id}/`
+</context>
+
+<invariants>
+1. **Skills invoked DIRECTLY in-context** — coordinator runs `$skill {resolved_args}` itself, sequentially. NO spawn_agents_on_csv, NO wave/CSV/worker.
+2. **Coordinator owns the loop** — classify → decompose → resolve chain → for each step: resolve args → invoke skill → read result → persist → next.
+3. **Decomposition contract shared with maestro-ralph** — broad/lifecycle intents run S_DECOMPOSE producing the SAME additive block (`boundary_contract`, `execution_criteria`, `task_decomposition`). Reference maestro-ralph `A_DECOMPOSE_TASKS`
+4. **Goal is tool-created** — `A_DECOMPOSE_TASKS` calls `create_goal` with sub-goal success criteria. `update_goal` on convergence; held while aborted/paused
+5. **status.json 唯一真源** — 不生成 `goal-checklist.md`；step 含 `command_scope` + `command_path` + `completion_confirmed`
+6. **Topology awareness** — chain catalog 含 grill / brainstorm / blueprint / analyze-macro(text) / analyze(numeric) / roadmap / plan(三路径) / execute / ...
+6.5. **Grill is interactive-only** — auto_mode MUST skip grill stage and route directly to brainstorm; grill requires Socratic Q&A with the user
+7. **D-007 milestone 反查** — 数字 phase 步骤的 `milestone_id` 由 `state.json.milestones[].phase_slugs` 反查
+8. **schema 向后兼容** — decomposition 字段可选；`steps[]` 由 post-goal-audit 动态生长（goal_ref tagged）；既有字段不删不改；`waves` 保留空数组
+9. **Sequential execution** — one step at a time in index order; each step's result read before the next starts
+10. **Abort on failure** — failed step → mark remaining skipped → report (goal stays bound for `--continue`)
+</invariants>
+
+<state_machine>
+
+<states>
+S_PARSE         — 解析参数、检测 flags              PERSIST: —
+S_CONTINUE      — 加载已有 session，定位 resume 点   PERSIST: session (loaded)
+S_CLASSIFY      — 意图分类、解析 chain (A_CLASSIFY)   PERSIST: —
+S_DECOMPOSE     — 边界澄清、写执行准则+子目标、建 goal PERSIST: session.boundary_contract, .execution_criteria, .task_decomposition
+S_CREATE        — 创建 session + status.json         PERSIST: session.status, session.steps[]
+S_DRY_RUN       — 显示 chain 后结束                  PERSIST: —
+S_CONFIRM       — 用户确认（auto_mode 跳过）          PERSIST: —
+S_STEP_LOOP     — 逐步直接调用 skill → 读结果 → 循环  PERSIST: session.current_step, session.steps[], session.context
+S_DECISION_EVAL — 评估 post-goal-audit 决策节点       PERSIST: —
+S_COMPLETE      — 标记完成、释放目标                  PERSIST: session.status = "completed"
+S_ABORTED       — 失败中止、标记剩余 skipped          PERSIST: session.status = "aborted"
+S_FALLBACK      — 意图无法分类，请求输入              PERSIST: —
+</states>
+
+<transitions>
+
+S_PARSE:
+  → S_CONTINUE    WHEN: --continue flag
+  → S_CLASSIFY    WHEN: intent text present
+  → S_FALLBACK    WHEN: no intent AND no flags
+
+S_CONTINUE:
+  → S_STEP_LOOP   WHEN: session found, has pending steps     DO: A_RESUME_SESSION
+  → S_FALLBACK    WHEN: no session found
+
+S_CLASSIFY:
+  → S_DECOMPOSE   WHEN: chain resolved                      DO: A_CLASSIFY
+  → S_FALLBACK    WHEN: no match AND auto_mode
+  → S_CLASSIFY    WHEN: no match AND not auto_mode          DO: A_CLARIFY_INTENT
+                   GUARD: max 1 clarification attempt → S_FALLBACK
+
+S_DECOMPOSE:
+  → S_CREATE      DO: A_DECOMPOSE_TASKS
+                   GUARD: broad intent (重构/全面/重写/迁移/overhaul/migrate/rewrite) on multi-step lifecycle chain → MUST clarify even if auto_mode
+                   GUARD: single-step chain OR narrow intent OR chain ∈ {status,init,quick} → skip decomposition (pass through)
+
+S_CREATE:
+  → S_DRY_RUN     WHEN: --dry-run flag                      DO: A_CREATE_SESSION
+  → S_CONFIRM     WHEN: not auto_mode                       DO: A_CREATE_SESSION
+  → S_STEP_LOOP   WHEN: auto_mode                           DO: A_CREATE_SESSION
+
+S_DRY_RUN:
+  → END           DO: display chain with step types + sub-goal summary
+
+S_CONFIRM:
+  → S_STEP_LOOP   WHEN: user confirms
+  → S_ABORTED     WHEN: user cancels
+
+S_STEP_LOOP:
+  → S_DECISION_EVAL WHEN: next step.type == "decision"
+  → S_STEP_LOOP   WHEN: next step.type == "skill"           DO: A_EXEC_STEP
+  → S_COMPLETE    WHEN: no pending steps
+  → S_ABORTED     WHEN: step failed (auto_mode: retry once then abort)
+
+S_DECISION_EVAL:                                            ENTRY: A_GOAL_AUDIT_EVALUATE (produces verdict)
+  → S_STEP_LOOP   WHEN: verdict == all_met                  DO: A_APPLY_GOAL_DONE
+  → S_STEP_LOOP   WHEN: verdict == has_unmet                DO: A_APPLY_GOAL_FIX
+  → S_ABORTED     WHEN: retry >= max_retries AND unmet      DO: escalate (insert quality-debug "{gaps}")
+
+S_COMPLETE:
+  → END           DO: A_FINALIZE
+
+S_ABORTED:
+  → END           DO: A_ABORT_REPORT
+
+S_FALLBACK:
+  → S_CLASSIFY    WHEN: user provides new intent            DO: request_user_input
+  → END           WHEN: user cancels
+
+</transitions>
+
+<actions>
+
+### A_CLASSIFY
+
+**Layer 1: Exact-match (fast path)**
+- `--chain <name>` flag → validate against chainMap, use directly (E002 if not found)
+- `continue`/`next`/`go`/`继续`/`下一步` → `state_continue`
+- `status`/`状态`/`dashboard` → `status`
+
+If matched, skip to chain resolution.
+
+**Layer 2: Semantic intent matching**
+
+Directly match user intent to the best `task_type` (maps to chain in Chain Map). Use LLM semantic understanding — no rigid keyword lookup.
+
+Extract:
+```json
+{
+  "task_type": "<from chain catalog below>",
+  "scope":     "<module/file/area or null>",
+  "issue_id":  "<ISS-XXXXXXXX-NNN if mentioned, else null>",
+  "phase_ref": "<integer if mentioned, else null>",
+  "urgency":   "<low|normal|high>"
+}
+```
+
+**Chain catalog — select by best semantic fit:**
+
+| task_type | When user intent is about... |
+|-----------|---------------------------|
+| `grill` | Stress-test, challenge assumptions, Socratic questioning on a plan/idea (**skip when auto_mode — grill is interactive-only**) |
+| `quick` | Simple/small task, add a feature, quick change |
+| `blueprint` | Formal spec generation (Product Brief / PRD / Architecture / Epics) |
+| `analyze_macro` | Broad/medium intent w/o numeric phase — explore impact, produce scope_verdict |
+| `plan_from_analyze` | Plan directly from analyze artifact (no roadmap, scope=standalone) |
+| `plan_from_blueprint` | Plan directly from blueprint artifact (scope=standalone) |
+| `plan` | Plan, design, architect a phase |
+| `execute` | Implement, develop, code a phase |
+| `analyze` | Understand, investigate, evaluate code (numeric phase) |
+| `verify` | Check goals met, validate results (routes to quality-review) |
+| `review` | Code quality review |
+| `test` | Run or create tests, UAT |
+| `test_gen` | Generate tests for coverage gaps |
+| `debug` | Diagnose, troubleshoot, fix broken behavior |
+| `refactor` | Restructure, clean up, reduce tech debt |
+| `init` | Initialize project |
+| `sync` | Update/sync documentation |
+| `retrospective` | Phase review, post-mortem, 复盘 |
+| `learn` | Capture insights, record learnings |
+| `release` | Publish, ship, tag version |
+| `amend` | Revise workflow commands |
+| `compose` | Design/compose reusable workflows |
+| `overlay` | Create/edit command overlays |
+| `update` | Update maestro itself |
+| `harvest` | Extract knowledge from artifacts |
+| `domain_add` | Register a domain term into glossary |
+| `domain_list` | List registered domain terms |
+| `domain_discover` | Discover domain term candidates from codebase |
+| `wiki` | Manage wiki graph |
+| `knowhow` | Manage knowhow entries |
+| `ui_design` | UI design, build new UI |
+| `issue` | Issue CRUD — create, list, close, query |
+| `issue_discover` | Discover/find issues in codebase |
+| `issue_analyze` | Analyze a specific issue |
+| `issue_plan` | Plan fix for an issue |
+| `issue_execute` | Fix issue end-to-end (auto-upgrades to issue-full) |
+| `feature` | Standard feature: plan→execute→review |
+| `full-lifecycle` | Complete phase: plan→execute→review→test→audit→complete |
+| `brainstorm-driven` | Start from exploration/brainstorm |
+| `spec-driven` | From spec/requirements (heavy, with init) |
+| `roadmap-driven` | From requirements (light, with init) |
+| `analyze-plan-execute` | Fast track: analyze→plan→execute |
+| `execute-review` | Resume after planning |
+| `review-fix` | Fix review-blocked issues |
+| `quality-loop` | Full quality improvement cycle |
+| `quality-loop-partial` | Partial quality fix |
+| `quality-fix` | Analyze gaps→plan→execute→review |
+| `deploy` | Verify then release |
+| `milestone-close` | Close/transition milestone |
+| `milestone-release` | Release milestone with version tag |
+| `phase_transition` | Transition phase: audit→complete |
+| `next-milestone` | Advance to next milestone |
+| `state_continue` | Continue from current project state |
+
+**Selection priorities:**
+1. `issue_id` present → prefer issue chains
+2. UI/design/界面/页面/原型 → prefer `ui_design`
+3. 正式规格/spec-generate/7-phase → `blueprint` (single-step) 或 `blueprint-driven`
+4. 压力测试/拷问/grill/stress-test → `grill` (single-step); **auto_mode → skip grill, route to `brainstorm-driven` instead**
+5. 头脑风暴/探索 → `brainstorm-driven`
+5. Broad/medium intent + 无数字 phase → `analyze_macro`（产 scope_verdict）；后续 large→roadmap链；medium/small→`plan_from_analyze`
+6. 已有 analyze artifact 直达 plan → `plan_from_analyze`
+7. 已有 blueprint artifact 直达 plan → `plan_from_blueprint`
+8. Multiple lifecycle steps implied → prefer multi-step chains
+9. Single specific action → prefer single-step chains
+10. "问题" describing broken behavior → `debug`; tracked item with ISS-ID → `issue`
+11. Simple task, no lifecycle context → `quick`
+12. Global fallback → `quick`
+
+**Clarity scoring**: 3=task_type+scope+phase, 2=task_type+scope, 1=task_type only, 0=empty.
+If `clarity < 2` and not `auto_mode` → transition to A_CLARIFY_INTENT.
+
+**Layer 4: State-based routing** (when `taskType === 'state_continue'`)
+
+Read `.workflow/state.json` and route by condition:
+
+| Condition | Chain |
+|-----------|-------|
+| Not initialized | `init` |
+| No phases, no roadmap, has accumulated_context | `next-milestone` |
+| No phases | `brainstorm-driven` |
+| pending + has context | `plan` |
+| pending, no context | `analyze` |
+| exploring/planning + has plan | `execute-review` |
+| exploring/planning, no plan | `plan` |
+| executing, all tasks done | `review` |
+| executing, tasks remain | `execute` |
+| reviewing, verdict == BLOCK | `review-fix` |
+| reviewing, verdict != BLOCK + UAT pending | `test` |
+| reviewing, verdict != BLOCK + UAT passed | `milestone-close` |
+| reviewing, verdict != BLOCK + UAT failed | `debug` |
+| testing, UAT passed | `milestone-close` |
+| testing, UAT not passed | `debug` |
+| completed | `milestone-close` |
+| blocked | `debug` |
+| fallback | `status` |
+
+**Chain resolution order:**
+1. `forceChain` → `chainMap[forceChain]` (E002 if not found)
+2. `state_continue` → Layer 4 state routing → `{ chain, argsOverride? }`
+3. `taskToChain[taskType]` → alias lookup (see Chain Aliases below)
+4. `chainMap[taskType]` → direct lookup
+
+**Phase resolution**: structured extraction `phase_ref` → fallback regex (`phase N` or bare number) → `projectState.current_phase`.
+
+### A_CLARIFY_INTENT
+
+1. `request_user_input` with available chain types
+2. Re-classify with user response
+
+### A_DECOMPOSE_TASKS
+
+与 maestro-ralph `A_DECOMPOSE_TASKS` 共享分解契约。Condensed:
+
+1. 分类意图广度。narrow / 单步 / `{status,init,quick}` 链跳过
+2. broad/medium → `request_user_input` ≤3 轮：Scope / Constraints / Definition of Done
+3. 派生 `execution_criteria` + `task_decomposition`（每个 sub-goal 含 `done_when` + `evidence` + `lifecycle` + `completion_confirmed: false`）
+4. **status.json 唯一真源**：写入 `boundary_contract` / `execution_criteria` / `task_decomposition`；不生成 markdown 清单
+5. 链路末尾（evidence 产出步骤后、milestone-complete/close-out 前）追加 `decision:post-goal-audit`。S_DECISION_EVAL 据此动态生长 `steps[]`
+6. **Register goal via `create_goal`:**
+   ```
+   create_goal({ objective: "Maestro {chain}: {intent} — converge {N} sub-goals within boundary",
+     success_criteria: task_decomposition.map(g => `${g.id}: ${g.done_when}`),
+     constraints: [...execution_criteria, "stay within boundary_contract; resume via $maestro --continue"] })
+   ```
+
+### A_CREATE_SESSION
+
+1. Read `.workflow/state.json` 获取 phase / milestone（D-007 反查 `phase_slugs`）；读最新 macro analyze artifact 注入 `scope_verdict` + `analyze_macro_id`；读最新 blueprint artifact 注入 `blueprint_id`
+2. Resolve chain's skill list from Chain Map (see appendix)
+3. **Prevalidate via `Bash("maestro ralph skills --platform codex --json --quiet")`** 一次性拉取所有可用 codex skills（global `~/.codex/skills/` + project `.codex/skills/`，project 覆盖 global），匹配 skill 名得到：
+   - 命中 → `command_scope = "global" | "project"`，`command_path = <绝对 SKILL.md 路径>`
+   - 未命中 → `command_scope = "missing"`, `command_path = null`
+4. Create `.workflow/.maestro/maestro-{YYYYMMDD-HHMMSS}/status.json`（与 ralph 共用 schema）:
+   ```json
+   {
+     "session_id", "source": "maestro", "intent", "task_type", "chain_name",
+     "phase", "phase_is_new": false, "milestone": "",
+     "scope_verdict": null, "analyze_macro_id": null, "blueprint_id": null,
+     "auto_mode": false,
+     "context": { "issue_id": null, "scratch_dir": null, "plan_dir": null,
+       "analysis_dir": null, "brainstorm_dir": null, "blueprint_dir": null, "grill_id": null },
+     "steps": [{
+       "index": 0, "type": "skill|decision",
+       "skill": "", "args": "",
+       "stage": "", "scope": null,
+       "command_scope": "global|project|missing|null",
+       "command_path": "~/.codex/skills/{name}/SKILL.md | .codex/skills/{name}/SKILL.md | null",
+       "milestone_id": null, "source_artifact_ref": null,
+       "status": "pending", "goal_ref": null,
+       "completion_confirmed": false, "completion_status": null,
+       "completion_evidence": null, "completed_at": null
+     }],
+     "waves": [], "current_step": 0, "status": "running",
+     "boundary_contract": {}, "execution_criteria": [],
+     "task_decomposition": [{ "id": "G1", "goal": "", "done_when": "", "evidence": "",
+       "status": "pending|done", "completion_confirmed": false, "completed_at": null }],
+     "task_decomposition_all_done": false
+   }
+   ```
+   Decomposition fields written ONLY if A_DECOMPOSE_TASKS produced them (additive)
+5. Validate: 所有 step 的 `command_scope != "missing"`；否则 raise E006 列出缺失 skill
+6. Initialize tracking:
+   - If decomposed: goal already registered by A_DECOMPOSE_TASKS. Else: `create_goal({ objective: "Maestro {chain}: {N} steps [{skill list}]" })`
+   - `update_plan({ plan: steps.map(step => ({ step, status: "pending" })) })`
+
+### A_RESUME_SESSION
+
+1. Glob `.workflow/.maestro/maestro-*/status.json` sorted desc, load most recent
+2. Find first pending step → set as resume point
+3. Rebuild `update_plan` from status.json (completed→"completed", current→"in_progress", rest→"open")
+
+### A_EXEC_STEP
+
+Direct in-context skill invocation — **replaces the old spawn/wave/CSV mechanism**.
+
+1. **buildSkillCall**: replace placeholders + append auto-yes flag if `auto_mode` (see Appendix: Auto-Yes Flag Map):
+
+   | Placeholder | Source |
+   |-------------|--------|
+   | `{phase}` | session.phase |
+   | `{plan_dir}` | session.context.plan_dir |
+   | `{analysis_dir}` | session.context.analysis_dir |
+   | `{brainstorm_dir}` | session.context.brainstorm_dir |
+   | `{spec_session_id}` | session.context.spec_session_id |
+   | `{GRL}` | session.context.grill_id |
+   | `{ANL}` | session.analyze_macro_id |
+   | `{BLP}` | session.blueprint_id |
+   | `{intent}` | session.intent |
+
+   **--from auto-injection**: 当 step 是 `maestro-plan`，args 含 `{phase}` 但无 `--from` 且无 `--dir`，且 `session.context.analysis_dir` 已填充 → 查 state.json 同 phase+milestone 最新 completed analyze artifact → 注入 `--from analyze:{id}`，写 `step.source_artifact_ref`
+2. Mark step `status="running"`, persist status.json + `update_plan` (this step → in_progress)
+3. **Invoke the skill directly**: execute `$skill {resolved_args}` in coordinator context (NO spawn). Read its produced artifacts directly
+4. On success: capture summary; mark step `status="done"`. **Barrier-context update** (when step is a context-producing skill):
+   | Skill | Read | Context Updates |
+   |-------|------|-----------------|
+   | maestro-grill | grill-report.md, state.json | grill_id |
+   | maestro-analyze | context.md, state.json | analysis_dir, gaps, phase |
+   | maestro-plan | plan.json, .task/TASK-*.json | plan_dir, task_count |
+   | maestro-brainstorm | .brainstorming/ | brainstorm_dir, features |
+   | maestro-roadmap | specs/ | spec_session_id |
+   | maestro-execute | results.csv | exec_completed, exec_failed |
+5. On failure: mark `status="failed"`; auto_mode → retry once → still failed → S_ABORTED
+6. Persist status.json + `update_plan` after every step
+
+### A_GOAL_AUDIT_EVALUATE
+
+S_DECISION_EVAL 入口；镜像 maestro-ralph `A_GOAL_AUDIT_EVALUATE`。Condensed:
+
+1. 读 `session.task_decomposition`（status.json，真源）
+2. 对每个 `status != "done"` 的子目标：解析 `evidence` 产物
+3. `maestro delegate --role analyze --mode analysis` 读取 evidence、对照 done_when 判定，返回 `STATUS=all_met|has_unmet / UNMET=[{id,gap,target_phase}] / CONFIDENCE_SCORE`
+4. status.json 为写入目标：每个达成子目标 `status="done"` + `completed_at=now`；然后从 status.json 重渲染 checklist（Sync Rule）
+5. Verdict（`all_met` / `has_unmet`）由 S_DECISION_EVAL 消费。GUARD: retry >= max_retries AND still unmet → escalate
+
+### A_APPLY_GOAL_FIX
+
+**Dynamic step-growth core** (mirrors maestro-ralph). For each unmet sub-goal (grouped by target_phase), insert before the post-goal-audit node a scoped mini-loop `$maestro-plan --gaps {phase} "G{n}: {gap}" → $maestro-execute {phase}`, each tagged `goal_ref: "G{n}"`, type `"skill"`. Re-append `decision:post-goal-audit {retry+1}`. Reindex, increment retry, persist + `update_plan`. `steps[]` grew.
+
+### A_APPLY_GOAL_DONE
+
+1. status.json：全部 `task_decomposition[*].status="done"` + `completion_confirmed=true` + `completed_at=now` + `task_decomposition_all_done=true`
+2. `update_goal({ status: "complete" })`
+3. 继续到 chain 的终结步骤
+
+### A_FINALIZE
+
+1. Set `session.status = "completed"`, write status.json
+2. Sync `update_plan`: all steps → "completed"
+3. `update_goal({ status: "complete" })` — release goal (idempotent if already released)
+4. Generate completion report (see Appendix: Report Format)
+
+### A_ABORT_REPORT
+
+1. Mark remaining steps `skipped` in status.json
+2. Set `session.status = "aborted"`, write status.json; sync `update_plan`
+3. Do NOT call `update_goal` — goal stays for `--continue` resume
+4. Display abort report with failure details
+
+</actions>
+
+</state_machine>
+
+<appendix>
+
+### Chain Map (Full)
+
+**Single-step chains:**
+
+| Chain | Command + Args |
+|-------|---------------|
+| `grill` | `maestro-grill "{intent}"` |
+| `status` | `manage-status` |
+| `init` | `maestro-init` |
+| `blueprint` | `maestro-blueprint "{intent}"` |
+| `analyze_macro` | `maestro-analyze "{intent}"` |
+| `analyze` | `maestro-analyze {phase}` |
+| `ui_design` | `maestro-impeccable build "{phase}"` |
+| `plan` | `maestro-plan {phase}` |
+| `plan_from_analyze` | `maestro-plan --from analyze:{analyze_macro_id}` |
+| `plan_from_blueprint` | `maestro-plan --from blueprint:{blueprint_id}` |
+| `execute` | `maestro-execute {phase}` |
+| `verify` | `quality-review {phase}` |
+| `test_gen` | `quality-auto-test {phase}` |
+| `auto_test` | `quality-auto-test {phase}` |
+| `test` | `quality-test {phase}` |
+| `debug` | `quality-debug "{description}"` |
+| `integration_test` | `quality-auto-test {phase}` |
+| `refactor` | `quality-refactor "{description}"` |
+| `review` | `quality-review {phase}` |
+| `retrospective` | `quality-retrospective {phase}` |
+| `learn` | `maestro-learn "{description}"` |
+| `sync` | `quality-sync` |
+| `milestone_audit` | `maestro-milestone-audit` |
+| `milestone_complete` | `maestro-milestone-complete` |
+| `codebase_rebuild` | `manage-codebase-rebuild` |
+| `codebase_refresh` | `manage-codebase-refresh` |
+| `spec_setup` | `spec-setup` |
+| `spec_add` | `spec-add "{description}"` |
+| `spec_load` | `spec-load` |
+| `spec_map` | `manage-codebase-rebuild` |
+| `spec_remove` | `spec-remove "{description}"` |
+| `domain_add` | `domain-add "{description}"` |
+| `domain_list` | `domain-list` |
+| `domain_discover` | `domain-discover` |
+| `knowhow_capture` | `manage-knowhow-capture "{description}"` |
+| `knowhow` | `manage-knowhow "{description}"` |
+| `issue` | `manage-issue "{description}"` |
+| `issue_discover` | `manage-issue-discover "{description}"` |
+| `issue_analyze` | `maestro-analyze --gaps "{description}"` |
+| `issue_plan` | `maestro-plan --gaps` |
+| `issue_execute` | `maestro-execute` |
+| `quick` | `maestro-quick "{description}"` |
+| `harvest` | `manage-harvest "{description}"` |
+| `wiki` | `manage-wiki` |
+| `wiki_connect` | `wiki-connect` |
+| `wiki_digest` | `wiki-digest` |
+| `business_test` | `quality-auto-test {phase}` |
+| `amend` | `maestro-amend "{description}"` |
+| `release` | `maestro-milestone-release` |
+| `compose` | `maestro-composer "{description}"` |
+| `play` | `maestro-player "{description}"` |
+| `update` | `maestro-update` |
+| `overlay` | `maestro-overlay "{description}"` |
+| `link_coordinate` | `maestro-link-coordinate "{description}"` |
+
+**Multi-step chains:**
+
+| Chain | Steps (→ = sequential, [B] = context-producing barrier) |
+|-------|---------------------------------------|
+| `feature` | [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `quality-fix` | [B] maestro-analyze --gaps → [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
+| `deploy` | quality-review → maestro-milestone-release |
+| `blueprint-driven` | maestro-init → [B] maestro-blueprint → [B] maestro-plan --from blueprint:{BLP} → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `analyze-macro-driven` | [B] maestro-analyze "{intent}" → ◆ post-analyze-scope → (large: [B] maestro-roadmap --from analyze:{ANL} → [B] maestro-analyze {phase} → [B] maestro-plan {phase}) / (medium\|small: [B] maestro-plan --from analyze:{ANL}) → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `grill-brainstorm` | [B] maestro-grill → [B] maestro-brainstorm --from grill:{GRL} → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto (**auto_mode: skip grill step, fall back to brainstorm-driven**) |
+| `brainstorm-driven` | [B] maestro-brainstorm → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `ui-craft-build` | maestro-impeccable build → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `roadmap-driven` | maestro-init → [B] maestro-roadmap → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `next-milestone` | [B] maestro-roadmap → [B] maestro-plan → [B] maestro-execute → quality-review → manage-harvest --auto |
+| `full-lifecycle` | [B] maestro-plan → [B] maestro-execute → quality-review → quality-test → maestro-milestone-audit → maestro-milestone-complete → manage-harvest --auto |
+| `execute-review` | [B] maestro-execute → quality-review |
+| `analyze-plan-execute` | [B] maestro-analyze -q → [B] maestro-plan --dir {scratch_dir} → [B] maestro-execute --dir {scratch_dir} → manage-harvest --auto |
+| `quality-loop` | quality-review → quality-test → quality-debug --from-uat → [B] maestro-plan --gaps → [B] maestro-execute |
+| `quality-loop-partial` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
+| `review-fix` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review |
+| `milestone-close` | maestro-milestone-audit → maestro-milestone-complete |
+| `milestone-release` | maestro-milestone-audit → maestro-milestone-release |
+| `phase_transition` | maestro-milestone-audit → maestro-milestone-complete |
+| `issue-full` | [B] maestro-analyze --gaps → [B] maestro-plan --gaps → [B] maestro-execute → quality-review → manage-issue close → manage-harvest --auto |
+| `issue-quick` | [B] maestro-plan --gaps → [B] maestro-execute → quality-review → manage-issue close |
+
+> When S_DECOMPOSE ran, a `decision:post-goal-audit` node is appended as the final node (after the last evidence-producing step; before milestone-complete/close-out if the chain ends with one). `[B]` now denotes a context-producing skill (artifacts read into `session.context`) — execution is still sequential (no parallelism; spawning removed).
+
+**Chain Aliases** (taskType → chain):
+
+| taskType | Chain |
+|----------|-------|
+| `spec_generate` | `blueprint-driven` |
+| `spec-driven` | `blueprint-driven` |
+| `brainstorm` | `brainstorm-driven` |
+| `issue_execute` | `issue-full` |
+| `analyze_macro` | `analyze-macro-driven` |
+
+### Auto-Yes Flag Map
+
+| Skill | Flag |
+|-------|------|
+| maestro-init, maestro-analyze, maestro-brainstorm, maestro-blueprint, maestro-impeccable, maestro-roadmap | `-y` |
+| maestro-plan, maestro-execute, maestro-milestone-complete | `-y` |
+| quality-auto-test, quality-retrospective | `-y` |
+| quality-test | `-y --auto-fix` |
+
+### Context-Producing Skills
+
+`maestro-analyze`, `maestro-plan`, `maestro-brainstorm`, `maestro-roadmap`, `maestro-execute` — their artifacts are read into `session.context` after the step completes (see A_EXEC_STEP step 4). Other skills produce no coordinator context. No parallelism — all steps run sequentially.
+
+### Error Codes
+
+| Code | Severity | Condition | Recovery |
+|------|----------|-----------|----------|
+| E001 | error | Intent unclassifiable after clarification | Default to `feature` chain |
+| E002 | error | Intent unresolvable after retry | List chains, abort |
+| E003 | error | Step skill invocation failed | auto_mode retry once, then abort chain |
+| E004 | error | Context artifact not found | Retry step once, then abort |
+| E005 | error | --continue: no session found | List sessions, prompt |
+| E006 | error | command_scope == "missing" for one or more steps | List missing skills, abort build |
+| W001 | warning | Context artifact partial | Continue with available context |
+
+### Success Criteria
+
+- [ ] Intent classified and chain resolved
+- [ ] Chain catalog 覆盖 blueprint / analyze_macro / plan_from_analyze / plan_from_blueprint / blueprint-driven / analyze-macro-driven 等新拓扑路径
+- [ ] D-007: 数字 phase 步骤的 `milestone_id` 通过 `state.json.milestones[].phase_slugs` 反查；写入 step
+- [ ] plan step args 支持 `{phase}` / `--from analyze:{ANL_ID}` / `--from blueprint:{BLP_ID}` 三路径，`source_artifact_ref` 写入
+- [ ] Broad lifecycle intents decomposed (≤3 boundary questions); narrow/single-step skip
+- [ ] Goal registered via built-in `create_goal`; status.json decomposition fields additive-only
+- [ ] status.json 唯一真源；无 markdown 清单
+- [ ] 每个 step 含 `command_scope` + `command_path` + `completion_confirmed` 字段
+- [ ] post-goal-audit node appended as final node; unmet sub-goals dynamically grow steps[] (goal_ref tagged)
+- [ ] Session dir initialized with status.json before first step
+- [ ] Every skill invoked DIRECTLY in-context — NO spawn_agents_on_csv, NO wave/CSV/worker
+- [ ] Sequential execution; status.json + update_plan persisted after every step
+- [ ] Context-producing skills' artifacts read into session.context before next step's args assembled
+- [ ] Failed step → remaining skipped → abort reported (goal held for --continue)
+- [ ] --dry-run shows chain + sub-goal summary, no execution
+- [ ] --continue resumes from first pending step
+- [ ] update_goal released on convergence (A_APPLY_GOAL_DONE / A_FINALIZE); held while aborted
+
+### Report Format
+
+```
+=== MAESTRO COMPLETE ===
+Session:  {sessionId}
+Chain:    {chain}
+Steps:    {completed}/{total}   Sub-goals: {done}/{total}
+
+STEP RESULTS:
+  [1] $maestro-analyze --gaps  →  ✓  found 3 gaps
+  [2] $maestro-plan --gaps     →  ✓  12 tasks
+  [◆] post-goal-audit          →  ✓  all sub-goals met
+  ...
+
+State:    .workflow/.maestro/{sessionId}/status.json
+Resume:   $maestro --continue
+```
+
+</appendix>