maestro-flow 0.5.36 → 0.5.38

package/.claude/commands/odyssey-planex.md

@@ -1,608 +1,608 @@
----
-name: odyssey-planex
-description: Requirement-driven iterative cycle — plan, execute, strict verify, fix loop until acceptance criteria met
-argument-hint: "<requirement> [--max-iterations N] [--skip-generalize] [--auto] [--method agent|cli|auto] [--executor <tool>] [--skip-verify] [-y] [-c]"
-allowed-tools:
-  - Read
-  - Write
-  - Edit
-  - Bash
-  - Glob
-  - Grep
-  - Agent
-  - AskUserQuestion
----
-<purpose>
-Requirement-to-delivery closed loop: parse requirement → define acceptance criteria →
-plan → execute → verify → fix gaps → iterate until ALL criteria pass.
-</purpose>
-
-<boundary>
-**范围内:** 单一需求的实现闭环 — 从需求解析到验收标准全部通过 + 泛化同类场景
-**范围外:** 多需求编排 → `/maestro-roadmap` | 深度 debugging → `/odyssey-debug` | 代码审查 → `/odyssey-review-test-fix` | UI 优化 → `/odyssey-ui`
-**探索自由度:** 边界内自由探索 — 可自主分解任务、选择实现策略、迭代修复。verify→fix 循环内可尝试不同方案。
-**Zero-residual principle:** Every failing criterion MUST be fixed or explicitly escalated with specific reason. "Close enough" is not passing. "Pre-existing gap" is not a valid skip reason — if within scope, address it.
-**模板支持:** `--template <name>` 从预定义需求模板启动，自动生成匹配的验收标准和任务分解：
-
-| Template | 预设 criteria 模式 | 适用场景 |
-|----------|-------------------|---------|
-| `feature` | 用户故事验收 + 边界测试 + UI 验证 | 新功能开发 |
-| `bugfix` | 回归测试 + 根因确认 + 边界覆盖 | Bug 修复 |
-| `refactor` | 行为不变验证 + 性能基准 + API 兼容 | 重构 |
-| `migration` | 数据一致性 + 回滚验证 + 性能对比 | 数据/API 迁移 |
-| `api-endpoint` | 请求/响应契约 + 错误处理 + 权限校验 | API 开发 |
-</boundary>
-
-<execution_discipline>
-**三条铁律（所有阶段适用）:**
-
-1. **Phase auto-commit** — 每个阶段完成后**自动** `git commit`，无需用户确认
-   - 代码变更 + understanding.md → `git add` → `git commit -m "odyssey-planex({slug}): {phase} — {摘要}"`
-   - session.json / evidence.ndjson 为运行时状态，不纳入 commit
-
-2. **Confident edits only, but must attempt** — only modify what you're confident about; record decisions only when genuinely requiring human judgment
-   - Confident → edit code directly, commit
-   - Needs decision → record `evidence.ndjson {"phase":"decision","status":"pending"}`, don't touch code
-   - No speculative changes
-   - ⚠️ **Decision gate** — ONLY these qualify as decisions (not fixes):
-     - Cross-module architectural tradeoffs requiring human direction
-     - Ambiguous business semantics where the fix could alter intended behavior
-     - Requires new dependency or breaking API change
-   - ❌ "Unsure how to fix", "Large scope", "Pre-existing issue" are NOT valid decision reasons — either fix it, or explain specifically why it's unfixable
-
-3. **多 CLI 辅助** — 利用 `maestro delegate` 调用多个 CLI 工具交叉验证
-   - 计划阶段: `--role analyze` 获取任务分解建议
-   - 修复前后: `--role review` 确认无回归
-   - verify 阶段: cli-review 类型标准自动 delegate
-
-4. **禁止以上下文消耗为由中断** — harness 自动处理 context compression，以"上下文不足"或"已执行 N 个阶段"为由中断属于纪律违反；必须完整走完状态机直到 S_RECORD → END
-</execution_discipline>
-
-<context>
-$ARGUMENTS — requirement description and optional flags.
-
-**Flags:**
-| Flag | Description | Default |
-|------|-------------|---------|
-| `--template <name>` | 预定义需求模板 | — |
-| `--max-iterations N` | Max verify→fix cycles before escalation | 3 |
-| `--skip-generalize` | Skip S_GENERALIZE + S_DISCOVER | false |
-| `--auto` | CLI delegate calls without confirmation | false |
-| `--method agent\|cli\|auto` | Execution method: Agent tool, CLI delegate, or auto-select | `auto` |
-| `--executor <tool>` | Explicit executor tool for CLI delegate mode | First enabled in config |
-| `--skip-verify` | Skip execution post-validation gate | false |
-| `-y` | Auto-confirm — decisions recorded as `deferred` | false |
-| `-c` | Resume most recent session | — |
-
-**Session**: `SESSION_DIR = .workflow/scratch/{YYYYMMDD}-planex-odyssey-{slug}/`
-
-**Output — 3 files:**
-```
-SESSION_DIR/
-  ├── session.json       # state + criteria + iterations + plan
-  ├── evidence.ndjson    # append-only log (phase distinguishes origin)
-  └── understanding.md   # evolving narrative (8 sections, one per phase)
-```
-
-**session.json schema:**
-```json
-{
-  "session_id": "planex-odyssey-{YYYYMMDD-HHmmss}",
-  "requirement": "",
-  "flags": { "max_iterations": 3, "skip_generalize": false, "auto": false, "auto_confirm": false },
-  "current_state": "S_INTAKE",
-  "acceptance_criteria": [
-    {"id":"AC1","criterion":"","verify_method":"test|grep|cli-review|manual","status":"pending","evidence":"","passed_at":null}
-  ],
-  "plan": { "tasks": [{"id":"T1","title":"","description":"","criteria_refs":["AC1"],"status":"pending","files_modified":[],"domain":"general","executor":"agent"}], "created_at":"" },
-  "execution_config": {
-    "method": "auto",
-    "default_executor": "",
-    "domain_routing": { "frontend": "", "backend": "", "default": "agent" },
-    "code_review_tool": "Skip",
-    "verification_tool": "Auto",
-    "confirmed": false
-  },
-  "iterations": [
-    {"iteration":1,"started_at":"","completed_at":"","criteria_before":{"passed":0,"total":0},"criteria_after":{"passed":0,"total":0},"gaps_fixed":[],"files_modified":[]}
-  ],
-  "current_iteration": 0,
-  "patterns": [
-    {"id":"P1","source":"AC1 fix","layer":"syntax|semantic|structural","signature":"","description":"","risk":"","fix_template":""}
-  ],
-  "generalization_stats": {"patterns_extracted":0,"total_hits":0,"cross_layer_confirmed":0,"by_layer":{"syntax":0,"semantic":0,"structural":0},"deepening_triggered":false},
-  "phase_goals": [],
-  "phase_goals_all_done": false,
-  "self_iteration_log": [],
-  "cross_phase_loops": 0, "max_loops": 5,
-  "created_at": "", "updated_at": ""
-}
-```
-
-**evidence.ndjson** — one JSON per line, `phase` field = `planning|execution|verification|fix|decision|generalization|discovery|self-iteration`
-
-**understanding.md sections:** §1 Requirement & Criteria ← S_INTAKE, §2 Plan ← S_PLAN, §3 Execution ← S_EXECUTE, §4 Verification (per iter) ← S_VERIFY, §5 Fix Log (per iter) ← S_FIX, §6 Generalization ← S_GENERALIZE, §7 Discoveries ← S_DISCOVER, §8 Learnings ← S_RECORD
-
-**phase_goals[]:**
-| ID | Goal | Done When | Phase | Skip When |
-|----|------|-----------|-------|-----------|
-| G1 | Acceptance criteria defined | ≥1 criterion in acceptance_criteria[] | S_INTAKE | — |
-| G2 | Plan created | session.json.plan populated | S_PLAN | — |
-| G3 | Implementation complete | all plan tasks executed | S_EXECUTE | — |
-| G4 | All criteria pass | all acceptance_criteria[].status == passed | S_VERIFY | — |
-| G5 | Pattern generalized | patterns[] populated ≥1 entry | S_GENERALIZE | skip_generalize |
-| G6 | Discoveries triaged | all scan hits classified | S_DISCOVER | skip_generalize |
-| G7 | Learnings persisted | spec entries created OR no actionable | S_RECORD | — |
-
-### Pre-load（可选，缺失不阻塞）
-
-| 层级 | 命令 | 作用 |
-|------|------|------|
-| Codebase docs | Read `.workflow/codebase/ARCHITECTURE.md` | 模块边界，架构约束 |
-| Wiki search | `maestro search "<requirement keywords>" --json` | 先前实现、相关决策（取 top 5） |
-| Coding specs | `maestro spec load --category coding` | 编码规范 + 可发现的 knowhow 工具 |
-| UI specs（条件） | 若涉及前端 → `maestro spec load --category ui` | UI 规范 |
-| Role knowledge | `maestro search --category coding` → 选相关 → `maestro wiki load <id>` | 累积实现领域知识 |
-| Prior sessions | `Glob(".workflow/scratch/*-planex-odyssey-*")` | 相关 odyssey 会话 |
-
-### Knowledge Persistence（S_RECORD 中写入产出文件）
-
-S_RECORD 阶段将可沉淀知识 **写入 understanding.md §8 Learnings**，按以下分类结构化：
-
-| 分类 | 写入内容 | 后续建议命令 |
-|------|---------|-------------|
-| 多轮 fix cycle pattern | 问题场景 + fix 迭代过程 + 最终方案 | `/spec-add debug "..."` |
-| 可复用实现模式 | 模式描述 + 适用场景 + 代码模板 | `/spec-add coding "..."` |
-| 验收标准模板 | 标准模板 + verify_method 建议 | `/spec-add review "..."` |
-| 泛化 pattern | pattern 签名 + 风险说明 + fix 模板 | `/spec-add coding "..."` |
-
-**两步模式：** 执行中写入产出文件（临时记录）→ 任务完成后用户通过 next_step_routing 沉淀为永久知识。执行过程中不调用外部 Skill。
-</context>
-
-<self_iteration>
-**Quality Gate** — auto-evaluate after each analytical stage. Insufficient → re-enter with expanded strategy.
-
-| Dimension | Sufficient | Insufficient |
-|-----------|-----------|-------------|
-| Coverage | All known files/modules analyzed | Missed targets discoverable via grep/git log |
-| Depth | ≥80% findings have file:line evidence | Most findings lack specifics |
-| Actionability | Each conclusion has concrete next action | Only vague "consider" recommendations |
-
-**Rules:** stage complete → evaluate 3 dims → any insufficient → re-enter (max **3 rounds** per stage). Record to evidence.ndjson `{"phase":"self-iteration","type":"quality-gate","stage":"S_XXX","round":N,"assessment":{...},"expansion":"strategy"}`.
-
-**Expansion:** Round 1 = broaden scope (more dirs, more delegate angles). Round 2 = shift perspective (different CLI tool, reverse-trace from expected result). Round 3 = combine both + targeted deep-dive on remaining gaps.
-
-**Applies to:** S_PLAN, S_VERIFY, S_GENERALIZE
-</self_iteration>
-
-<state_machine>
-
-<states>
-S_INTAKE      — parse requirement, define acceptance criteria       PERSIST: session.json + understanding.md §1
-S_PLAN        — decompose tasks, generate execution plan            PERSIST: session.json.plan + evidence (planning) + understanding.md §2
-S_EXECUTE     — implement tasks                                     PERSIST: code + evidence (execution) + understanding.md §3
-S_VERIFY      — iron gate: check every acceptance criterion         PERSIST: evidence (verification) + understanding.md §4
-S_FIX         — targeted fix for failing criteria (loops to VERIFY) PERSIST: code + evidence (fix) + understanding.md §5
-S_GENERALIZE  — extract patterns, 4-agent scan                     PERSIST: session.json.patterns + understanding.md §6
-S_DISCOVER    — triage scan hits, route decisions                   PERSIST: evidence (discovery|decision) + understanding.md §7
-S_RECORD      — persist learnings, final summary                   PERSIST: understanding.md §8 + spec entries
-</states>
-
-<transitions>
-S_INTAKE → S_INTAKE  WHEN -c + session found (resume)
-S_INTAKE → S_PLAN    WHEN requirement + criteria defined
-S_INTAKE → S_INTAKE  WHEN no requirement → AskUserQuestion
-
-S_PLAN → S_EXECUTE
-S_EXECUTE → S_VERIFY
-
-S_VERIFY → S_GENERALIZE  WHEN all passed AND NOT skip_generalize
-S_VERIFY → S_RECORD      WHEN all passed AND skip_generalize
-S_VERIFY → S_FIX         WHEN some failed AND iteration < max
-S_VERIFY → S_PLAN        WHEN fundamental plan flaw discovered, loops < max_loops → cross_phase_loops++ (重规划)
-S_VERIFY → S_RECORD      WHEN some failed AND iteration >= max (escalate)
-
-S_FIX → S_VERIFY (loop)
-
-S_GENERALIZE → S_DISCOVER  WHEN hits found
-S_GENERALIZE → S_RECORD    WHEN no hits
-
-S_DISCOVER → S_EXECUTE     : discovery finds area needing same implementation → cross_phase_loops++
-S_DISCOVER → S_RECORD      : triage complete AND remaining_actionable == 0
-S_DISCOVER → S_RECORD      : loops >= max_loops → MUST log each unfixed item with specific reason (blanket "pre-existing" is forbidden)
-
-S_RECORD → END
-</transitions>
-
-<actions>
-
-### A_INTAKE
-
-1. Parse requirement and flags, generate slug, create SESSION_DIR
-2. **Define acceptance criteria** — analyze requirement → derive testable criteria. Each gets `verify_method`: test | grep | cli-review | manual
-   - **Normal**: AskUserQuestion to confirm/edit
-   - **`-y`**: auto-derive, record `{"phase":"decision","type":"criteria-confirmation","status":"deferred"}`
-3. Search prior knowledge: `maestro search`, related sessions
-4. Write session.json + understanding.md §1. Mark G1 done. Display Goal Prompt (see Appendix)
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): INTAKE — 目标解析"`
-
-### A_PLAN
-
-1. Decompose requirement into ordered tasks mapped to acceptance criteria
-2. CLI-assisted planning (optional):
-   ```bash
-   maestro delegate "PURPOSE: Create implementation plan for: {requirement}
-   TASK: Decompose into subtasks | Map to acceptance criteria | Identify dependencies
-   MODE: analysis
-   CONTEXT: @**/* | Criteria: {criteria_summary}
-   EXPECTED: JSON [{task_id, title, description, criteria_refs, deps}]
-   " --role analyze --mode analysis
-   ```
-   Execute with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
-3. Write session.json.plan, append evidence (planning), update understanding.md §2. Mark G2 done.
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): PLAN — 计划制定"`
-
-### A_EXECUTE
-
-#### Step 1: Execution Options Confirmation
-
-**Skip if** `-y` flag OR `--method` explicitly set OR `execution_config.confirmed == true` (resume).
-
-Load available tools: `maestro delegate-config show --json` → extract enabled tools and domain tags.
-
-```
-AskUserQuestion({
-  questions: [
-    {
-      question: "任务如何执行？选择一种方式，或 Other 指定域路由规则（如 '前端gemini 后端codex 其余agent'）",
-      header: "Executor",
-      options: [
-        { label: "Auto (Recommended)", description: "域路由: frontend→{frontendTool}, backend→{backendTool}, general→agent" },
-        { label: "Agent", description: "Claude Code Agent 执行所有任务（最快）" },
-        // 每个 enabled CLI tool 一个选项
-        ...availableTools.map(t => ({ label: t, description: `${t} CLI 执行所有任务` }))
-      ]
-    },
-    {
-      question: "执行后运行代码审查？",
-      header: "Review",
-      options: [
-        { label: "Skip", description: "不审查" },
-        ...availableTools.map(t => ({ label: `${t} Review`, description: `${t} CLI: git diff 质量审查` }))
-      ]
-    },
-    {
-      question: "验证门控？（外部模型检查收敛 + 结构 + 反模式）",
-      header: "Verify",
-      options: [
-        { label: "Auto (Recommended)", description: `Delegate 到 ${availableTools[0]} 做收敛+结构+反模式检查` },
-        ...availableTools.map(t => ({ label: t, description: `${t}: 验证门控` })),
-        { label: "Skip", description: "不验证" }
-      ]
-    }
-  ]
-})
-```
-
-Parse response → write `execution_config` to session.json, set `confirmed: true`.
-
-`--skip-verify` flag overrides verification to `"Skip"`.
-
-#### Step 2: Executor Resolution
-
-Per-task domain routing (when method == "auto"):
-
-| Domain | Keywords / Patterns | File Extensions |
-|--------|-------------------|-----------------|
-| frontend | UI, component, page, style, layout, CSS, view | .tsx/.jsx/.vue/.css/.html/.svelte |
-| backend | API, server, database, service, algorithm, worker | .go/.rs/.java/.py/.sql/.proto |
-| general | mixed, config, tests, unclear | .ts/.js/other |
-
-Resolution: `execution_config.domain_routing[domain]` → fallback `domain_routing.default` ("agent").
-
-Log routing per task:
-```
-T1 [frontend] → gemini
-T2 [backend]  → codex
-T3 [general]  → agent
-```
-
-#### Step 3: Task Execution
-
-Execute tasks per plan order. Independent tasks (no cross-dependency) may run in parallel.
-
-**Agent path:**
-```
-Spawn Agent with:
-  task definition, acceptance criteria refs, prior task summaries, specs_content
-Agent implements → verifies convergence criteria → auto-fix (max 3) → returns result
-```
-
-**CLI path (via maestro delegate):**
-```bash
-maestro delegate "PURPOSE: Implement task ${task_id}: ${title}; success = criteria ${criteria_refs} satisfied
-TASK: ${description} | Read existing code first | Verify convergence criteria after changes
-MODE: write
-CONTEXT: @${scope}/**/* | Criteria: ${criteria_summary}
-EXPECTED: Working code changes, convergence evidence, summary of what was done
-CONSTRAINTS: Scope limited to task files | Follow project specs
-
-## Acceptance Criteria (must satisfy)
-${criteria_refs.map(ref => criteria[ref].criterion).join('\n')}
-
-## Implementation Steps
-${task.description}
-
-## Project Specs
-${specs_content}
-
-## Prior Task Summaries
-${prior_summaries}
-" --to ${resolved_executor} --mode write --id planex-${slug}-${task_id}
-```
-
-Run CLI delegate with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
-
-**Deviation Rule** — max 3 auto-fix attempts per task:
-1. First attempt: normal dispatch
-2. Retry: `--resume planex-${slug}-${task_id}` with simplified prompt
-3. Final: fallback to Agent path
-4. All 3 fail → mark task `blocked`, record checkpoint, continue remaining tasks
-
-#### Step 4: Per-Task Evidence
-
-Per completed task:
-- Record evidence: `{"phase":"execution","type":"task-completed","task_id":"T1","executor":"agent|gemini|...","files_modified":[],"summary":"","attempt":1}`
-- Update task status in session.json plan
-
-#### Step 5: Post-Execution Validation
-
-**Skip if** `execution_config.verification_tool == "Skip"` OR `--skip-verify` OR no completed tasks.
-
-**Check 1: Summary Consistency** — cross-check task status vs actual file changes (git diff).
-
-**Check 2: CLI Verification Gate** — delegate to external model:
-```bash
-maestro delegate "PURPOSE: Verify execution output meets acceptance criteria; success = all criteria verified with file:line evidence
-TASK:
-1. CONVERGENCE: For each criterion, read actual code, verify behavior exists, report status with evidence
-2. EXISTENCE: Verify all expected files exist on disk
-3. SUBSTANCE: Verify real implementation — flag stubs, placeholders, TODO-only
-4. ANTI-PATTERNS: Scan for TODO/FIXME/HACK, console.log debug, disabled tests
-MODE: analysis
-CONTEXT: @${modified_files}
-EXPECTED: JSON { convergence: [{criterion, status, evidence}], issues: [{type, file, line, severity}], overall: passed|gaps_found }
-CONSTRAINTS: Read-only | Check ALL criteria exhaustively | Evidence must be file:line
-
-## Acceptance Criteria (verify each)
-${acceptance_criteria.map(c => c.criterion).join('\n')}
-
-## Modified Files
-${modified_files.join('\n')}
-" --to ${execution_config.verification_tool} --mode analysis
-```
-
-Execute with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
-
-On result:
-- `overall == "passed"` → proceed to S_VERIFY (criteria gate) with boosted confidence
-- `overall == "gaps_found"` → log findings, proceed to S_VERIFY (criteria will catch failures)
-
-**Check 3: Code Review** (if `execution_config.code_review_tool != "Skip"`):
-```bash
-maestro delegate "Review git diff for correctness, style, bugs" --to ${code_review_tool} --mode analysis --rule analysis-review-code-quality
-```
-
-#### Step 6: Completion
-
-Update understanding.md §3. Mark G3 done.
-
-📌 **Auto-commit**: `git add -A && git commit -m "odyssey-planex({slug}): EXECUTE — 实现执行"`
-
-### A_VERIFY
-
-Iron gate — every acceptance criterion checked objectively.
-
-**Verify each criterion by method:**
-| Method | Action |
-|--------|--------|
-| `test` | Run relevant tests, check pass/fail |
-| `grep` | Grep for expected pattern |
-| `cli-review` | `maestro delegate --role review --mode analysis` with criterion as focus |
-| `manual` | **Normal**: AskUserQuestion / **`-y`**: record `deferred` |
-
-Record per criterion: `{"phase":"verification","type":"criterion-check","criterion_id":"AC1","method":"","result":"passed|failed","evidence":"","iteration":N}`. Update acceptance_criteria[].status. Append to iterations[].
-
-Update understanding.md §4 with pass/fail table.
-
-**Route:** all passed → mark G4 done → next state. Some failed + iteration < max → S_FIX. Some failed + iteration >= max → **Normal**: AskUserQuestion (continue/lower bar/accept) / **`-y`**: `deferred`, proceed S_RECORD.
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): VERIFY — 验收验证"`
-
-### A_FIX
-
-1. Increment current_iteration
-2. For each failed criterion: diagnose gap → targeted code fix
-3. CLI fix review (optional):
-   ```bash
-   maestro delegate "PURPOSE: Review fixes for failing criteria
-   TASK: Check fix correctness | Verify no regressions on passing criteria
-   MODE: analysis
-   CONTEXT: @{modified_files} | Passing: {passing} | Fixed: {fixed}
-   EXPECTED: JSON {verdict, regression_risk, concerns}
-   " --role review --mode analysis
-   ```
-4. Append evidence (fix), update understanding.md §5 → S_VERIFY
-
-📌 **Auto-commit**: `git add -A && git commit -m "odyssey-planex({slug}): FIX — 修复"`
-
-### A_GENERALIZE
-
-Extract reusable patterns from implementation, scan codebase for similar sites.
-
-**Pattern extraction (3 layers):**
-| Layer | Method | Example |
-|-------|--------|---------|
-| Syntax | Code regex patterns | validation/error handling patterns |
-| Semantic | Logic pattern description | missing similar checks at other entry points |
-| Structural | File/module structure match | sibling modules lacking same treatment |
-
-**4-agent parallel scan** (spawn 4 Agents):
-| Agent | Strategy | Scope |
-|-------|----------|-------|
-| Syntax grep | Grep syntax-layer signatures | full project |
-| Semantic scan | Check for same anti-pattern in related modules | related modules |
-| Structural match | Find structurally similar files | full project |
-| Historical grep | `git log -S "{pattern}"` | full git history |
-
-Each returns: `[{pattern_id, file, line, context, risk_level, layer, confidence}]`
-
-**Cross-layer dedup:** multi-layer hit on same file:line → boost confidence. Historical hit with existing fix → `already_handled`. Single layer only → `needs_review`.
-
-**Quality Gate** (self-iteration) → evaluate coverage/depth/actionability.
-
-Write understanding.md §6, generalization_stats. Mark G5 done.
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): GENERALIZE — 泛化扫描"`
-
-### A_DISCOVER
-
-1. **Triage:** per hit, read context (+-10 lines), classify as `already_handled` | `needs_treatment` | `low_risk`
-2. **Route:**
-   | Classification | Normal | `-y` |
-   |---------------|--------|------|
-   | needs_treatment | AskUserQuestion: create issue / plan next iter | auto create issue, `deferred` |
-   | low_risk | Record only | Record only |
-   | already_handled | Skip | Skip |
-3. Append evidence (discovery + decision), update understanding.md §7. Mark G6 done.
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): DISCOVER — 发现分类"`
-
-### A_RECORD
-
-1. Finalize understanding.md §8 — iteration summary, what worked, what needed rework
-2. Write learnings to understanding.md §8: 按 Knowledge Persistence 表分类记录（临时），completion summary 列出建议的 `/spec-add` 命令
-3. Pending decisions: **Normal** → AskUserQuestion. **`-y`** → display deferred count.
-4. Goal audit: check all phase_goals[*].completion_confirmed. Mark G7 done.
-5. Output completion summary:
-   ```
-   --- PLANEX ODYSSEY COMPLETE ---
-   Requirement: {requirement}
-   Criteria:    {passed}/{total} passed
-   Iterations:  {N} cycles
-   Patterns:    {patterns_extracted} ({by_layer} distribution)
-   Scan hits:   {total_hits} ({cross_layer_confirmed} cross-layer confirmed)
-   Issues:      {N} created | Decisions: {N} resolved, {M} pending, {K} deferred
-   Learnings:   {N} spec entries
-   Self-iter:   {N} rounds across {M} stages
-   Goals:       {done}/{total} confirmed ({skipped} skipped)
-   Status:      {ALL_PASSED|PARTIAL|ESCALATED}
-   ---
-   ```
-
-📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): RECORD — 会话总结"`
-
-</actions>
-
-<appendix>
-
-### Goal Prompt Template
-
-**⚠️ 时机守卫：仅在 A_INTAKE 完成后显示一次（session 创建后、开始 Plan 前）。A_RECORD 完成时禁止重新显示。**
-
-```
-📋 Planex Odyssey 会话已创建。可随时复制以下 /goal 设定终止条件（执行过程中输入即可）：
-
-/goal 完成以下目标：
-{for each G in phase_goals where status != "skipped":}
-- {G.id}: {G.goal} — 完成条件: {G.done_when}
-{end for}
-验收标准：
-{for each AC in acceptance_criteria:}
-- {AC.id}: {AC.criterion} (验证方式: {AC.verify_method})
-{end for}
-穷尽迭代：直到 acceptance_criteria[*] 全部 status==passed
-且 phase_goals_all_done=true 才停。verify 失败自动 fix→re-verify 循环。
-每轮修复后重新验证，新发现的标准违反继续修，不超过 max_iterations。
-遇到 phase=decision 的 pending 必须 AskUserQuestion，不得自行 resolve。
-不允许"接近通过"，验收标准必须 ALL pass。
-```
-
-完成时仅输出 completion summary，不重复此提示。
-
-### `-y` Auto-Confirm Behavior
-
-| Decision Point | Normal | `-y` |
-|----------------|--------|------|
-| S_INTAKE criteria confirmation | AskUserQuestion | auto-derive, `deferred` |
-| S_EXECUTE execution options | AskUserQuestion (executor/review/verify) | use defaults (auto/Skip/Auto), `confirmed: true` |
-| S_EXECUTE task blocked (3 retries) | AskUserQuestion: continue or stop | auto continue, log blocked |
-| S_VERIFY manual criterion | AskUserQuestion | `deferred` |
-| S_VERIFY max iteration reached | AskUserQuestion | auto accept, `deferred` |
-| S_DISCOVER classification routing | AskUserQuestion | auto create issue, `deferred` |
-| S_DISCOVER ambiguous items | AskUserQuestion | all `deferred` |
-| S_RECORD decision list | AskUserQuestion | skip |
-| S_RECORD goal audit | AskUserQuestion | auto accept |
-
-### Iteration Model
-
-```
-S_EXECUTE → S_VERIFY ──all pass──→ S_GENERALIZE → S_DISCOVER → S_RECORD
-                │                       │
-           some fail + iter < max       no hits ─→ S_RECORD
-                ▼
-             S_FIX ──→ S_VERIFY (loop)
-```
-
-Max iterations (default 3) prevents infinite loops. Each iteration records criteria_before, gaps_fixed, criteria_after.
-
-</appendix>
-
-</state_machine>
-
-<error_codes>
-| Code | Severity | Condition | Recovery |
-|------|----------|-----------|----------|
-| E001 | error | No requirement provided | Provide requirement |
-| E003 | error | Resume but no session found | Start new |
-| E004 | error | Delegate failed | Retry or proceed without |
-| W001 | warning | No acceptance criteria derived | Manual definition needed |
-| W002 | warning | Max iterations reached, criteria still failing | Escalate to user |
-| W003 | warning | CLI review regression concern | Review before next iteration |
-| W004 | warning | Delegate parse failed | Use raw output |
-</error_codes>
-
-<success_criteria>
-- [ ] Requirement parsed and ≥1 acceptance criterion defined with verify_method
-- [ ] Plan created with tasks mapped to criteria
-- [ ] Execution options confirmed (executor/review/verify) before task dispatch
-- [ ] Tasks dispatched via resolved executor (agent/cli/auto domain routing)
-- [ ] Per-task deviation rule enforced (max 3 retries, fallback chain)
-- [ ] Post-execution validation gate run (unless --skip-verify)
-- [ ] Tasks executed with evidence logged (executor, attempt count, files_modified)
-- [ ] Every criterion verified by its method after each iteration
-- [ ] Failing criteria trigger targeted fix (not full re-implementation)
-- [ ] Iteration count tracked, max respected
-- [ ] **Every unfixed criterion has individual classification and reason** — blanket "pre-existing" labels are forbidden
-- [ ] understanding.md updated per phase (§1-§8)
-- [ ] Multi-layer generalization + 4-agent scan (unless --skip-generalize)
-- [ ] Discoveries classified and routed (unless --skip-generalize)
-- [ ] Quality Gate self-iteration triggered when insufficient, logged in self_iteration_log
-- [ ] phase_goals G1-G7 tracked and audited
-- [ ] Goal Prompt displayed once after intake
-- [ ] `-y` mode: no blocking prompts, deferred counted
-- [ ] Session resumable via -c
-- [ ] Completion summary with iteration stats
-</success_criteria>
-
-<next_step_routing>
-| Condition | Next step |
-|-----------|-----------|
-| All criteria passed | `/odyssey-review-test-fix <changed-files>` |
-| Max iterations, still failing | `/odyssey-debug "<failing criterion>"` |
-| Want formal review | `/quality-review <phase>` |
-| Issues from discoveries | `/manage-issue list --source planex-odyssey` |
-| Pattern worth documenting | `/learn-decompose <module>` |
-</next_step_routing>
+---
+name: odyssey-planex
+description: Requirement-driven iterative cycle — plan, execute, strict verify, fix loop until acceptance criteria met
+argument-hint: "<requirement> [--max-iterations N] [--skip-generalize] [--auto] [--method agent|cli|auto] [--executor <tool>] [--skip-verify] [-y] [-c]"
+allowed-tools:
+  - Read
+  - Write
+  - Edit
+  - Bash
+  - Glob
+  - Grep
+  - Agent
+  - AskUserQuestion
+---
+<purpose>
+Requirement-to-delivery closed loop: parse requirement → define acceptance criteria →
+plan → execute → verify → fix gaps → iterate until ALL criteria pass.
+</purpose>
+
+<boundary>
+**范围内:** 单一需求的实现闭环 — 从需求解析到验收标准全部通过 + 泛化同类场景
+**范围外:** 多需求编排 → `/maestro-roadmap` | 深度 debugging → `/odyssey-debug` | 代码审查 → `/odyssey-review-test-fix` | UI 优化 → `/odyssey-ui`
+**探索自由度:** 边界内自由探索 — 可自主分解任务、选择实现策略、迭代修复。verify→fix 循环内可尝试不同方案。
+**Zero-residual principle:** Every failing criterion MUST be fixed or explicitly escalated with specific reason. "Close enough" is not passing. "Pre-existing gap" is not a valid skip reason — if within scope, address it.
+**模板支持:** `--template <name>` 从预定义需求模板启动，自动生成匹配的验收标准和任务分解：
+
+| Template | 预设 criteria 模式 | 适用场景 |
+|----------|-------------------|---------|
+| `feature` | 用户故事验收 + 边界测试 + UI 验证 | 新功能开发 |
+| `bugfix` | 回归测试 + 根因确认 + 边界覆盖 | Bug 修复 |
+| `refactor` | 行为不变验证 + 性能基准 + API 兼容 | 重构 |
+| `migration` | 数据一致性 + 回滚验证 + 性能对比 | 数据/API 迁移 |
+| `api-endpoint` | 请求/响应契约 + 错误处理 + 权限校验 | API 开发 |
+</boundary>
+
+<execution_discipline>
+**三条铁律（所有阶段适用）:**
+
+1. **Phase auto-commit** — 每个阶段完成后**自动** `git commit`，无需用户确认
+   - 代码变更 + understanding.md → `git add` → `git commit -m "odyssey-planex({slug}): {phase} — {摘要}"`
+   - session.json / evidence.ndjson 为运行时状态，不纳入 commit
+
+2. **Confident edits only, but must attempt** — only modify what you're confident about; record decisions only when genuinely requiring human judgment
+   - Confident → edit code directly, commit
+   - Needs decision → record `evidence.ndjson {"phase":"decision","status":"pending"}`, don't touch code
+   - No speculative changes
+   - ⚠️ **Decision gate** — ONLY these qualify as decisions (not fixes):
+     - Cross-module architectural tradeoffs requiring human direction
+     - Ambiguous business semantics where the fix could alter intended behavior
+     - Requires new dependency or breaking API change
+   - ❌ "Unsure how to fix", "Large scope", "Pre-existing issue" are NOT valid decision reasons — either fix it, or explain specifically why it's unfixable
+
+3. **多 CLI 辅助** — 利用 `maestro delegate` 调用多个 CLI 工具交叉验证
+   - 计划阶段: `--role analyze` 获取任务分解建议
+   - 修复前后: `--role review` 确认无回归
+   - verify 阶段: cli-review 类型标准自动 delegate
+
+4. **禁止以上下文消耗为由中断** — harness 自动处理 context compression，以"上下文不足"或"已执行 N 个阶段"为由中断属于纪律违反；必须完整走完状态机直到 S_RECORD → END
+</execution_discipline>
+
+<context>
+$ARGUMENTS — requirement description and optional flags.
+
+**Flags:**
+| Flag | Description | Default |
+|------|-------------|---------|
+| `--template <name>` | 预定义需求模板 | — |
+| `--max-iterations N` | Max verify→fix cycles before escalation | 3 |
+| `--skip-generalize` | Skip S_GENERALIZE + S_DISCOVER | false |
+| `--auto` | CLI delegate calls without confirmation | false |
+| `--method agent\|cli\|auto` | Execution method: Agent tool, CLI delegate, or auto-select | `auto` |
+| `--executor <tool>` | Explicit executor tool for CLI delegate mode | First enabled in config |
+| `--skip-verify` | Skip execution post-validation gate | false |
+| `-y` | Auto-confirm — decisions recorded as `deferred` | false |
+| `-c` | Resume most recent session | — |
+
+**Session**: `SESSION_DIR = .workflow/scratch/{YYYYMMDD}-planex-odyssey-{slug}/`
+
+**Output — 3 files:**
+```
+SESSION_DIR/
+  ├── session.json       # state + criteria + iterations + plan
+  ├── evidence.ndjson    # append-only log (phase distinguishes origin)
+  └── understanding.md   # evolving narrative (8 sections, one per phase)
+```
+
+**session.json schema:**
+```json
+{
+  "session_id": "planex-odyssey-{YYYYMMDD-HHmmss}",
+  "requirement": "",
+  "flags": { "max_iterations": 3, "skip_generalize": false, "auto": false, "auto_confirm": false },
+  "current_state": "S_INTAKE",
+  "acceptance_criteria": [
+    {"id":"AC1","criterion":"","verify_method":"test|grep|cli-review|manual","status":"pending","evidence":"","passed_at":null}
+  ],
+  "plan": { "tasks": [{"id":"T1","title":"","description":"","criteria_refs":["AC1"],"status":"pending","files_modified":[],"domain":"general","executor":"agent"}], "created_at":"" },
+  "execution_config": {
+    "method": "auto",
+    "default_executor": "",
+    "domain_routing": { "frontend": "", "backend": "", "default": "agent" },
+    "code_review_tool": "Skip",
+    "verification_tool": "Auto",
+    "confirmed": false
+  },
+  "iterations": [
+    {"iteration":1,"started_at":"","completed_at":"","criteria_before":{"passed":0,"total":0},"criteria_after":{"passed":0,"total":0},"gaps_fixed":[],"files_modified":[]}
+  ],
+  "current_iteration": 0,
+  "patterns": [
+    {"id":"P1","source":"AC1 fix","layer":"syntax|semantic|structural","signature":"","description":"","risk":"","fix_template":""}
+  ],
+  "generalization_stats": {"patterns_extracted":0,"total_hits":0,"cross_layer_confirmed":0,"by_layer":{"syntax":0,"semantic":0,"structural":0},"deepening_triggered":false},
+  "phase_goals": [],
+  "phase_goals_all_done": false,
+  "self_iteration_log": [],
+  "cross_phase_loops": 0, "max_loops": 5,
+  "created_at": "", "updated_at": ""
+}
+```
+
+**evidence.ndjson** — one JSON per line, `phase` field = `planning|execution|verification|fix|decision|generalization|discovery|self-iteration`
+
+**understanding.md sections:** §1 Requirement & Criteria ← S_INTAKE, §2 Plan ← S_PLAN, §3 Execution ← S_EXECUTE, §4 Verification (per iter) ← S_VERIFY, §5 Fix Log (per iter) ← S_FIX, §6 Generalization ← S_GENERALIZE, §7 Discoveries ← S_DISCOVER, §8 Learnings ← S_RECORD
+
+**phase_goals[]:**
+| ID | Goal | Done When | Phase | Skip When |
+|----|------|-----------|-------|-----------|
+| G1 | Acceptance criteria defined | ≥1 criterion in acceptance_criteria[] | S_INTAKE | — |
+| G2 | Plan created | session.json.plan populated | S_PLAN | — |
+| G3 | Implementation complete | all plan tasks executed | S_EXECUTE | — |
+| G4 | All criteria pass | all acceptance_criteria[].status == passed | S_VERIFY | — |
+| G5 | Pattern generalized | patterns[] populated ≥1 entry | S_GENERALIZE | skip_generalize |
+| G6 | Discoveries triaged | all scan hits classified | S_DISCOVER | skip_generalize |
+| G7 | Learnings persisted | spec entries created OR no actionable | S_RECORD | — |
+
+### Pre-load（可选，缺失不阻塞）
+
+| 层级 | 命令 | 作用 |
+|------|------|------|
+| Codebase docs | Read `.workflow/codebase/ARCHITECTURE.md` | 模块边界，架构约束 |
+| Wiki search | `maestro search "<requirement keywords>" --json` | 先前实现、相关决策（取 top 5） |
+| Coding specs | `maestro load --type spec --category coding` | 编码规范 + 可发现的 knowhow 工具 |
+| UI specs（条件） | 若涉及前端 → `maestro load --type spec --category ui` | UI 规范 |
+| Role knowledge | `maestro search --category coding` → 选相关 → `maestro load --type knowhow --id <id>` | 累积实现领域知识 |
+| Prior sessions | `Glob(".workflow/scratch/*-planex-odyssey-*")` | 相关 odyssey 会话 |
+
+### Knowledge Persistence（S_RECORD 中写入产出文件）
+
+S_RECORD 阶段将可沉淀知识 **写入 understanding.md §8 Learnings**，按以下分类结构化：
+
+| 分类 | 写入内容 | 后续建议命令 |
+|------|---------|-------------|
+| 多轮 fix cycle pattern | 问题场景 + fix 迭代过程 + 最终方案 | `/spec-add debug "..."` |
+| 可复用实现模式 | 模式描述 + 适用场景 + 代码模板 | `/spec-add coding "..."` |
+| 验收标准模板 | 标准模板 + verify_method 建议 | `/spec-add review "..."` |
+| 泛化 pattern | pattern 签名 + 风险说明 + fix 模板 | `/spec-add coding "..."` |
+
+**两步模式：** 执行中写入产出文件（临时记录）→ 任务完成后用户通过 next_step_routing 沉淀为永久知识。执行过程中不调用外部 Skill。
+</context>
+
+<self_iteration>
+**Quality Gate** — auto-evaluate after each analytical stage. Insufficient → re-enter with expanded strategy.
+
+| Dimension | Sufficient | Insufficient |
+|-----------|-----------|-------------|
+| Coverage | All known files/modules analyzed | Missed targets discoverable via grep/git log |
+| Depth | ≥80% findings have file:line evidence | Most findings lack specifics |
+| Actionability | Each conclusion has concrete next action | Only vague "consider" recommendations |
+
+**Rules:** stage complete → evaluate 3 dims → any insufficient → re-enter (max **3 rounds** per stage). Record to evidence.ndjson `{"phase":"self-iteration","type":"quality-gate","stage":"S_XXX","round":N,"assessment":{...},"expansion":"strategy"}`.
+
+**Expansion:** Round 1 = broaden scope (more dirs, more delegate angles). Round 2 = shift perspective (different CLI tool, reverse-trace from expected result). Round 3 = combine both + targeted deep-dive on remaining gaps.
+
+**Applies to:** S_PLAN, S_VERIFY, S_GENERALIZE
+</self_iteration>
+
+<state_machine>
+
+<states>
+S_INTAKE      — parse requirement, define acceptance criteria       PERSIST: session.json + understanding.md §1
+S_PLAN        — decompose tasks, generate execution plan            PERSIST: session.json.plan + evidence (planning) + understanding.md §2
+S_EXECUTE     — implement tasks                                     PERSIST: code + evidence (execution) + understanding.md §3
+S_VERIFY      — iron gate: check every acceptance criterion         PERSIST: evidence (verification) + understanding.md §4
+S_FIX         — targeted fix for failing criteria (loops to VERIFY) PERSIST: code + evidence (fix) + understanding.md §5
+S_GENERALIZE  — extract patterns, 4-agent scan                     PERSIST: session.json.patterns + understanding.md §6
+S_DISCOVER    — triage scan hits, route decisions                   PERSIST: evidence (discovery|decision) + understanding.md §7
+S_RECORD      — persist learnings, final summary                   PERSIST: understanding.md §8 + spec entries
+</states>
+
+<transitions>
+S_INTAKE → S_INTAKE  WHEN -c + session found (resume)
+S_INTAKE → S_PLAN    WHEN requirement + criteria defined
+S_INTAKE → S_INTAKE  WHEN no requirement → AskUserQuestion
+
+S_PLAN → S_EXECUTE
+S_EXECUTE → S_VERIFY
+
+S_VERIFY → S_GENERALIZE  WHEN all passed AND NOT skip_generalize
+S_VERIFY → S_RECORD      WHEN all passed AND skip_generalize
+S_VERIFY → S_FIX         WHEN some failed AND iteration < max
+S_VERIFY → S_PLAN        WHEN fundamental plan flaw discovered, loops < max_loops → cross_phase_loops++ (重规划)
+S_VERIFY → S_RECORD      WHEN some failed AND iteration >= max (escalate)
+
+S_FIX → S_VERIFY (loop)
+
+S_GENERALIZE → S_DISCOVER  WHEN hits found
+S_GENERALIZE → S_RECORD    WHEN no hits
+
+S_DISCOVER → S_EXECUTE     : discovery finds area needing same implementation → cross_phase_loops++
+S_DISCOVER → S_RECORD      : triage complete AND remaining_actionable == 0
+S_DISCOVER → S_RECORD      : loops >= max_loops → MUST log each unfixed item with specific reason (blanket "pre-existing" is forbidden)
+
+S_RECORD → END
+</transitions>
+
+<actions>
+
+### A_INTAKE
+
+1. Parse requirement and flags, generate slug, create SESSION_DIR
+2. **Define acceptance criteria** — analyze requirement → derive testable criteria. Each gets `verify_method`: test | grep | cli-review | manual
+   - **Normal**: AskUserQuestion to confirm/edit
+   - **`-y`**: auto-derive, record `{"phase":"decision","type":"criteria-confirmation","status":"deferred"}`
+3. Search prior knowledge: `maestro search`, related sessions
+4. Write session.json + understanding.md §1. Mark G1 done. Display Goal Prompt (see Appendix)
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): INTAKE — 目标解析"`
+
+### A_PLAN
+
+1. Decompose requirement into ordered tasks mapped to acceptance criteria
+2. CLI-assisted planning (optional):
+   ```bash
+   maestro delegate "PURPOSE: Create implementation plan for: {requirement}
+   TASK: Decompose into subtasks | Map to acceptance criteria | Identify dependencies
+   MODE: analysis
+   CONTEXT: @**/* | Criteria: {criteria_summary}
+   EXPECTED: JSON [{task_id, title, description, criteria_refs, deps}]
+   " --role analyze --mode analysis
+   ```
+   Execute with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
+3. Write session.json.plan, append evidence (planning), update understanding.md §2. Mark G2 done.
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): PLAN — 计划制定"`
+
+### A_EXECUTE
+
+#### Step 1: Execution Options Confirmation
+
+**Skip if** `-y` flag OR `--method` explicitly set OR `execution_config.confirmed == true` (resume).
+
+Load available tools: `maestro delegate-config show --json` → extract enabled tools and domain tags.
+
+```
+AskUserQuestion({
+  questions: [
+    {
+      question: "任务如何执行？选择一种方式，或 Other 指定域路由规则（如 '前端agy 后端codex 其余agent'）",
+      header: "Executor",
+      options: [
+        { label: "Auto (Recommended)", description: "域路由: frontend→{frontendTool}, backend→{backendTool}, general→agent" },
+        { label: "Agent", description: "Claude Code Agent 执行所有任务（最快）" },
+        // 每个 enabled CLI tool 一个选项
+        ...availableTools.map(t => ({ label: t, description: `${t} CLI 执行所有任务` }))
+      ]
+    },
+    {
+      question: "执行后运行代码审查？",
+      header: "Review",
+      options: [
+        { label: "Skip", description: "不审查" },
+        ...availableTools.map(t => ({ label: `${t} Review`, description: `${t} CLI: git diff 质量审查` }))
+      ]
+    },
+    {
+      question: "验证门控？（外部模型检查收敛 + 结构 + 反模式）",
+      header: "Verify",
+      options: [
+        { label: "Auto (Recommended)", description: `Delegate 到 ${availableTools[0]} 做收敛+结构+反模式检查` },
+        ...availableTools.map(t => ({ label: t, description: `${t}: 验证门控` })),
+        { label: "Skip", description: "不验证" }
+      ]
+    }
+  ]
+})
+```
+
+Parse response → write `execution_config` to session.json, set `confirmed: true`.
+
+`--skip-verify` flag overrides verification to `"Skip"`.
+
+#### Step 2: Executor Resolution
+
+Per-task domain routing (when method == "auto"):
+
+| Domain | Keywords / Patterns | File Extensions |
+|--------|-------------------|-----------------|
+| frontend | UI, component, page, style, layout, CSS, view | .tsx/.jsx/.vue/.css/.html/.svelte |
+| backend | API, server, database, service, algorithm, worker | .go/.rs/.java/.py/.sql/.proto |
+| general | mixed, config, tests, unclear | .ts/.js/other |
+
+Resolution: `execution_config.domain_routing[domain]` → fallback `domain_routing.default` ("agent").
+
+Log routing per task:
+```
+T1 [frontend] → agy
+T2 [backend]  → codex
+T3 [general]  → agent
+```
+
+#### Step 3: Task Execution
+
+Execute tasks per plan order. Independent tasks (no cross-dependency) may run in parallel.
+
+**Agent path:**
+```
+Spawn Agent with:
+  task definition, acceptance criteria refs, prior task summaries, specs_content
+Agent implements → verifies convergence criteria → auto-fix (max 3) → returns result
+```
+
+**CLI path (via maestro delegate):**
+```bash
+maestro delegate "PURPOSE: Implement task ${task_id}: ${title}; success = criteria ${criteria_refs} satisfied
+TASK: ${description} | Read existing code first | Verify convergence criteria after changes
+MODE: write
+CONTEXT: @${scope}/**/* | Criteria: ${criteria_summary}
+EXPECTED: Working code changes, convergence evidence, summary of what was done
+CONSTRAINTS: Scope limited to task files | Follow project specs
+
+## Acceptance Criteria (must satisfy)
+${criteria_refs.map(ref => criteria[ref].criterion).join('\n')}
+
+## Implementation Steps
+${task.description}
+
+## Project Specs
+${specs_content}
+
+## Prior Task Summaries
+${prior_summaries}
+" --to ${resolved_executor} --mode write --id planex-${slug}-${task_id}
+```
+
+Run CLI delegate with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
+
+**Deviation Rule** — max 3 auto-fix attempts per task:
+1. First attempt: normal dispatch
+2. Retry: `--resume planex-${slug}-${task_id}` with simplified prompt
+3. Final: fallback to Agent path
+4. All 3 fail → mark task `blocked`, record checkpoint, continue remaining tasks
+
+#### Step 4: Per-Task Evidence
+
+Per completed task:
+- Record evidence: `{"phase":"execution","type":"task-completed","task_id":"T1","executor":"agent|agy|...","files_modified":[],"summary":"","attempt":1}`
+- Update task status in session.json plan
+
+#### Step 5: Post-Execution Validation
+
+**Skip if** `execution_config.verification_tool == "Skip"` OR `--skip-verify` OR no completed tasks.
+
+**Check 1: Summary Consistency** — cross-check task status vs actual file changes (git diff).
+
+**Check 2: CLI Verification Gate** — delegate to external model:
+```bash
+maestro delegate "PURPOSE: Verify execution output meets acceptance criteria; success = all criteria verified with file:line evidence
+TASK:
+1. CONVERGENCE: For each criterion, read actual code, verify behavior exists, report status with evidence
+2. EXISTENCE: Verify all expected files exist on disk
+3. SUBSTANCE: Verify real implementation — flag stubs, placeholders, TODO-only
+4. ANTI-PATTERNS: Scan for TODO/FIXME/HACK, console.log debug, disabled tests
+MODE: analysis
+CONTEXT: @${modified_files}
+EXPECTED: JSON { convergence: [{criterion, status, evidence}], issues: [{type, file, line, severity}], overall: passed|gaps_found }
+CONSTRAINTS: Read-only | Check ALL criteria exhaustively | Evidence must be file:line
+
+## Acceptance Criteria (verify each)
+${acceptance_criteria.map(c => c.criterion).join('\n')}
+
+## Modified Files
+${modified_files.join('\n')}
+" --to ${execution_config.verification_tool} --mode analysis
+```
+
+Execute with `run_in_background: true`, then wait for callback (do NOT halt the Odyssey flow).
+
+On result:
+- `overall == "passed"` → proceed to S_VERIFY (criteria gate) with boosted confidence
+- `overall == "gaps_found"` → log findings, proceed to S_VERIFY (criteria will catch failures)
+
+**Check 3: Code Review** (if `execution_config.code_review_tool != "Skip"`):
+```bash
+maestro delegate "Review git diff for correctness, style, bugs" --to ${code_review_tool} --mode analysis --rule analysis-review-code-quality
+```
+
+#### Step 6: Completion
+
+Update understanding.md §3. Mark G3 done.
+
+📌 **Auto-commit**: `git add -A && git commit -m "odyssey-planex({slug}): EXECUTE — 实现执行"`
+
+### A_VERIFY
+
+Iron gate — every acceptance criterion checked objectively.
+
+**Verify each criterion by method:**
+| Method | Action |
+|--------|--------|
+| `test` | Run relevant tests, check pass/fail |
+| `grep` | Grep for expected pattern |
+| `cli-review` | `maestro delegate --role review --mode analysis` with criterion as focus |
+| `manual` | **Normal**: AskUserQuestion / **`-y`**: record `deferred` |
+
+Record per criterion: `{"phase":"verification","type":"criterion-check","criterion_id":"AC1","method":"","result":"passed|failed","evidence":"","iteration":N}`. Update acceptance_criteria[].status. Append to iterations[].
+
+Update understanding.md §4 with pass/fail table.
+
+**Route:** all passed → mark G4 done → next state. Some failed + iteration < max → S_FIX. Some failed + iteration >= max → **Normal**: AskUserQuestion (continue/lower bar/accept) / **`-y`**: `deferred`, proceed S_RECORD.
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): VERIFY — 验收验证"`
+
+### A_FIX
+
+1. Increment current_iteration
+2. For each failed criterion: diagnose gap → targeted code fix
+3. CLI fix review (optional):
+   ```bash
+   maestro delegate "PURPOSE: Review fixes for failing criteria
+   TASK: Check fix correctness | Verify no regressions on passing criteria
+   MODE: analysis
+   CONTEXT: @{modified_files} | Passing: {passing} | Fixed: {fixed}
+   EXPECTED: JSON {verdict, regression_risk, concerns}
+   " --role review --mode analysis
+   ```
+4. Append evidence (fix), update understanding.md §5 → S_VERIFY
+
+📌 **Auto-commit**: `git add -A && git commit -m "odyssey-planex({slug}): FIX — 修复"`
+
+### A_GENERALIZE
+
+Extract reusable patterns from implementation, scan codebase for similar sites.
+
+**Pattern extraction (3 layers):**
+| Layer | Method | Example |
+|-------|--------|---------|
+| Syntax | Code regex patterns | validation/error handling patterns |
+| Semantic | Logic pattern description | missing similar checks at other entry points |
+| Structural | File/module structure match | sibling modules lacking same treatment |
+
+**4-agent parallel scan** (spawn 4 Agents):
+| Agent | Strategy | Scope |
+|-------|----------|-------|
+| Syntax grep | Grep syntax-layer signatures | full project |
+| Semantic scan | Check for same anti-pattern in related modules | related modules |
+| Structural match | Find structurally similar files | full project |
+| Historical grep | `git log -S "{pattern}"` | full git history |
+
+Each returns: `[{pattern_id, file, line, context, risk_level, layer, confidence}]`
+
+**Cross-layer dedup:** multi-layer hit on same file:line → boost confidence. Historical hit with existing fix → `already_handled`. Single layer only → `needs_review`.
+
+**Quality Gate** (self-iteration) → evaluate coverage/depth/actionability.
+
+Write understanding.md §6, generalization_stats. Mark G5 done.
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): GENERALIZE — 泛化扫描"`
+
+### A_DISCOVER
+
+1. **Triage:** per hit, read context (+-10 lines), classify as `already_handled` | `needs_treatment` | `low_risk`
+2. **Route:**
+   | Classification | Normal | `-y` |
+   |---------------|--------|------|
+   | needs_treatment | AskUserQuestion: create issue / plan next iter | auto create issue, `deferred` |
+   | low_risk | Record only | Record only |
+   | already_handled | Skip | Skip |
+3. Append evidence (discovery + decision), update understanding.md §7. Mark G6 done.
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): DISCOVER — 发现分类"`
+
+### A_RECORD
+
+1. Finalize understanding.md §8 — iteration summary, what worked, what needed rework
+2. Write learnings to understanding.md §8: 按 Knowledge Persistence 表分类记录（临时），completion summary 列出建议的 `/spec-add` 命令
+3. Pending decisions: **Normal** → AskUserQuestion. **`-y`** → display deferred count.
+4. Goal audit: check all phase_goals[*].completion_confirmed. Mark G7 done.
+5. Output completion summary:
+   ```
+   --- PLANEX ODYSSEY COMPLETE ---
+   Requirement: {requirement}
+   Criteria:    {passed}/{total} passed
+   Iterations:  {N} cycles
+   Patterns:    {patterns_extracted} ({by_layer} distribution)
+   Scan hits:   {total_hits} ({cross_layer_confirmed} cross-layer confirmed)
+   Issues:      {N} created | Decisions: {N} resolved, {M} pending, {K} deferred
+   Learnings:   {N} spec entries
+   Self-iter:   {N} rounds across {M} stages
+   Goals:       {done}/{total} confirmed ({skipped} skipped)
+   Status:      {ALL_PASSED|PARTIAL|ESCALATED}
+   ---
+   ```
+
+📌 **Auto-commit**: `git add understanding.md && git commit -m "odyssey-planex({slug}): RECORD — 会话总结"`
+
+</actions>
+
+<appendix>
+
+### Goal Prompt Template
+
+**⚠️ 时机守卫：仅在 A_INTAKE 完成后显示一次（session 创建后、开始 Plan 前）。A_RECORD 完成时禁止重新显示。**
+
+```
+📋 Planex Odyssey 会话已创建。可随时复制以下 /goal 设定终止条件（执行过程中输入即可）：
+
+/goal 完成以下目标：
+{for each G in phase_goals where status != "skipped":}
+- {G.id}: {G.goal} — 完成条件: {G.done_when}
+{end for}
+验收标准：
+{for each AC in acceptance_criteria:}
+- {AC.id}: {AC.criterion} (验证方式: {AC.verify_method})
+{end for}
+穷尽迭代：直到 acceptance_criteria[*] 全部 status==passed
+且 phase_goals_all_done=true 才停。verify 失败自动 fix→re-verify 循环。
+每轮修复后重新验证，新发现的标准违反继续修，不超过 max_iterations。
+遇到 phase=decision 的 pending 必须 AskUserQuestion，不得自行 resolve。
+不允许"接近通过"，验收标准必须 ALL pass。
+```
+
+完成时仅输出 completion summary，不重复此提示。
+
+### `-y` Auto-Confirm Behavior
+
+| Decision Point | Normal | `-y` |
+|----------------|--------|------|
+| S_INTAKE criteria confirmation | AskUserQuestion | auto-derive, `deferred` |
+| S_EXECUTE execution options | AskUserQuestion (executor/review/verify) | use defaults (auto/Skip/Auto), `confirmed: true` |
+| S_EXECUTE task blocked (3 retries) | AskUserQuestion: continue or stop | auto continue, log blocked |
+| S_VERIFY manual criterion | AskUserQuestion | `deferred` |
+| S_VERIFY max iteration reached | AskUserQuestion | auto accept, `deferred` |
+| S_DISCOVER classification routing | AskUserQuestion | auto create issue, `deferred` |
+| S_DISCOVER ambiguous items | AskUserQuestion | all `deferred` |
+| S_RECORD decision list | AskUserQuestion | skip |
+| S_RECORD goal audit | AskUserQuestion | auto accept |
+
+### Iteration Model
+
+```
+S_EXECUTE → S_VERIFY ──all pass──→ S_GENERALIZE → S_DISCOVER → S_RECORD
+                │                       │
+           some fail + iter < max       no hits ─→ S_RECORD
+                ▼
+             S_FIX ──→ S_VERIFY (loop)
+```
+
+Max iterations (default 3) prevents infinite loops. Each iteration records criteria_before, gaps_fixed, criteria_after.
+
+</appendix>
+
+</state_machine>
+
+<error_codes>
+| Code | Severity | Condition | Recovery |
+|------|----------|-----------|----------|
+| E001 | error | No requirement provided | Provide requirement |
+| E003 | error | Resume but no session found | Start new |
+| E004 | error | Delegate failed | Retry or proceed without |
+| W001 | warning | No acceptance criteria derived | Manual definition needed |
+| W002 | warning | Max iterations reached, criteria still failing | Escalate to user |
+| W003 | warning | CLI review regression concern | Review before next iteration |
+| W004 | warning | Delegate parse failed | Use raw output |
+</error_codes>
+
+<success_criteria>
+- [ ] Requirement parsed and ≥1 acceptance criterion defined with verify_method
+- [ ] Plan created with tasks mapped to criteria
+- [ ] Execution options confirmed (executor/review/verify) before task dispatch
+- [ ] Tasks dispatched via resolved executor (agent/cli/auto domain routing)
+- [ ] Per-task deviation rule enforced (max 3 retries, fallback chain)
+- [ ] Post-execution validation gate run (unless --skip-verify)
+- [ ] Tasks executed with evidence logged (executor, attempt count, files_modified)
+- [ ] Every criterion verified by its method after each iteration
+- [ ] Failing criteria trigger targeted fix (not full re-implementation)
+- [ ] Iteration count tracked, max respected
+- [ ] **Every unfixed criterion has individual classification and reason** — blanket "pre-existing" labels are forbidden
+- [ ] understanding.md updated per phase (§1-§8)
+- [ ] Multi-layer generalization + 4-agent scan (unless --skip-generalize)
+- [ ] Discoveries classified and routed (unless --skip-generalize)
+- [ ] Quality Gate self-iteration triggered when insufficient, logged in self_iteration_log
+- [ ] phase_goals G1-G7 tracked and audited
+- [ ] Goal Prompt displayed once after intake
+- [ ] `-y` mode: no blocking prompts, deferred counted
+- [ ] Session resumable via -c
+- [ ] Completion summary with iteration stats
+</success_criteria>
+
+<next_step_routing>
+| Condition | Next step |
+|-----------|-----------|
+| All criteria passed | `/odyssey-review-test-fix <changed-files>` |
+| Max iterations, still failing | `/odyssey-debug "<failing criterion>"` |
+| Want formal review | `/quality-review <phase>` |
+| Issues from discoveries | `/manage-issue list --source planex-odyssey` |
+| Pattern worth documenting | `/learn-decompose <module>` |
+</next_step_routing>