maestro-flow 0.5.46 → 0.5.48

package/.agents/skills/quality-auto-test/SKILL.md

@@ -54,8 +54,20 @@ Flags, artifact context resolution, and output formats defined in workflow auto-
    - Browse: `maestro search --category test`
    - Load task-relevant entries: `maestro load --type knowhow --id <id1> [id2...]`
 4. All are optional — proceed without if unavailable.
+
+**Output boundary**: ALL file writes MUST target `.workflow/scratch/{YYYYMMDD}-auto-test-P{N}-{slug}/` or `.workflow/state.json` only. Source code modifications are limited to test files generated by this command. NEVER modify existing production source code.
 </context>
 
+<invariants>
+1. **Test code isolation** — generated tests MUST be written to test directories following existing project conventions. NEVER modify production source code to make tests pass.
+2. **RED-GREEN integrity** — tests MUST be written to fail first (RED), then verified to pass against implementation (GREEN). NEVER write tests that pass trivially or assert nothing.
+3. **Layer ordering is strict** — L0→L3 execution is sequential and fail-fast. NEVER skip a layer or run L2 before L1 passes.
+4. **Existing tests are sacred** — NEVER modify, delete, or overwrite existing test files. New tests are additive only.
+5. **Convergence requires evidence** — pass rate and confidence score MUST be computed from actual execution results. NEVER infer or estimate pass rates without running tests.
+6. **Iteration has bounds** — outer iteration MUST NOT exceed `--max-iter` (default 5). Inner test_defect fix MUST NOT exceed 3 attempts per scenario before escalating.
+7. **Traceability is mandatory for spec source** — when route=spec, every test scenario MUST trace back to a REQ-*.md requirement. Untraceable tests are flagged.
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/auto-test.md' completely.
 
@@ -138,7 +150,7 @@ Append to state.json.artifacts[]:
 | E003 | error | No test framework detected | Install test framework or configure test runner |
 | W001 | warning | One or more test scenarios failed | Auto-iterate or suggest fix options |
 | W002 | warning | Max iterations reached without convergence | Review reflection-log.md, suggest debug |
-| W003 | warning | Degraded spec mode (no full spec package) | Consider running maestro-roadmap --mode full |
+| W003 | warning | Degraded spec mode (no full spec package) | Consider running /maestro-blueprint |
 </error_codes>
 
 <success_criteria>

package/.agents/skills/quality-debug/SKILL.md

@@ -50,8 +50,19 @@ Each artifact's type determines its outputs at `.workflow/{a.path}/`:
 - Role knowledge: `maestro search --category debug` → select relevant → `maestro load --type knowhow --id`
 
 **Output**: `DEBUG_DIR = .workflow/scratch/{YYYYMMDD}-debug-P{N}-{slug}/` (P{N} = phase number when phase-scoped; omit for standalone). Output directory rules defined in workflow debug.md Step 4.
+
+**Output boundary**: ALL file writes MUST target `DEBUG_DIR` or `.workflow/state.json` only. NEVER modify source code, test files, or other artifacts outside these paths during investigation.
 </context>
 
+<invariants>
+1. **Investigation is read-only on source** — debug MUST NOT modify source code, test files, or execution artifacts. Diagnosis produces understanding.md, evidence.ndjson, and reports only.
+2. **Root cause requires reproduction** — NEVER declare a root cause confirmed without evidence (code trace, log entry, or reproduction steps). Hypothesis without evidence stays "suspected".
+3. **Evidence is append-only** — evidence.ndjson entries MUST only be appended, NEVER modified or deleted. Each entry is an immutable observation.
+4. **Confidence is multi-factor** — root cause confidence MUST use the multi-factor scoring model (Step 7.0). NEVER use bare high/medium/low without dimensional breakdown.
+5. **Prior investigations are consulted** — when related debug artifacts exist, check understanding.md for prior root causes. NEVER re-investigate an already-confirmed root cause without new symptoms.
+6. **Fix direction is not fix execution** — debug produces `fix_direction` and `affected_files` but NEVER applies fixes. Fix execution belongs to plan→execute cycle.
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/debug.md' completely.
 

package/.agents/skills/quality-refactor/SKILL.md

@@ -38,8 +38,19 @@ If not provided, prompt user for scope.
    - Browse: `maestro search --category coding`
    - Identify task-relevant entries, then load: `maestro load --type knowhow --id <id1> [id2...]`
 4. All are optional — proceed without if unavailable.
+
+**Output boundary**: Refactoring modifies source files within the declared scope only. Ancillary outputs (reflection-log.md) MUST target `.workflow/scratch/{YYYYMMDD}-refactor-{slug}/`. NEVER modify files outside the confirmed scope without re-confirmation.
 </context>
 
+<invariants>
+1. **Plan before change** — NEVER apply refactoring changes without a confirmed plan. Every modification traces to a plan item.
+2. **Behavioral equivalence** — refactoring MUST preserve existing behavior. All tests MUST pass after each individual change, not just at the end.
+3. **Scope is locked after confirmation** — once the user confirms the refactoring plan, do NOT expand scope to include additional files or changes without re-confirmation.
+4. **Incremental verification** — each discrete refactoring step MUST be verified (tests run) before proceeding to the next. NEVER batch multiple unrelated changes into a single verification.
+5. **No feature creep** — refactoring MUST NOT add new functionality, change APIs, or alter public interfaces. If a beneficial API change is discovered, log it as a recommendation, do NOT apply it.
+6. **Rollback safety** — if any test fails after a refactoring step, revert that specific change before attempting alternatives. NEVER proceed with failing tests.
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/refactor.md' completely.
 

package/.agents/skills/quality-retrospective/SKILL.md

@@ -40,8 +40,19 @@ Arguments: $ARGUMENTS
 - `--lens <name>` — Restrict to specific lens (technical|process|quality|decision); default: all four
 
 Modes (scan/single/range/all) and storage paths defined in workflow retrospective.md Argument Shape and Stages 1-7.
+
+**Output boundary**: ALL file writes MUST target the phase's retrospective directory (`.workflow/scratch/{YYYYMMDD}-retrospective-P{N}/`), `.workflow/state.json`, `.workflow/issues.jsonl`, or `.workflow/specs/` (append-only). NEVER modify source code, verification.json, review.json, plan.json, or other existing artifacts.
 </context>
 
+<invariants>
+1. **Source artifacts are read-only** — NEVER modify verification.json, review.json, plan.json, or any execution artifact. Retrospective reads these for analysis only.
+2. **Stable insight IDs** — `INS-{8hex}` MUST be deterministic from `hash(phase_num + lens + title)`. Re-runs MUST NOT create duplicate insights.
+3. **Routing requires confirmation** — unless `-y` flag is set, every external write (issues.jsonl, spec entry, knowhow capture) MUST be confirmed by user before execution.
+4. **Lens independence** — each lens agent (technical/process/quality/decision) operates independently. One lens's findings MUST NOT suppress or override another's.
+5. **Append-only for specs** — learnings.md entries are appended as `<spec-entry>` blocks. NEVER overwrite or restructure existing entries.
+6. **Archive before overwrite** — if retrospective.json already exists for a phase, the existing file MUST be archived before writing a new version.
+</invariants>
+
 <execution>
 Follow `~/.maestro/workflows/retrospective.md` Stages 1–8 in order.
 

package/.agents/skills/quality-review/SKILL.md

@@ -56,8 +56,19 @@ Each artifact's type determines its outputs at `.workflow/{a.path}/`:
 - Role knowledge: `maestro search --category review` → select relevant → `maestro load --type knowhow --id`
 
 **Output**: `REVIEW_DIR = .workflow/scratch/{YYYYMMDD}-review-P{N}-{slug}/` (P{N} = phase number, enables directory-level identification as state.json fallback)
+
+**Output boundary**: ALL file writes MUST target `REVIEW_DIR` or `.workflow/state.json` only. NEVER modify source code, execution artifacts, or files outside these paths.
 </context>
 
+<invariants>
+1. **Review is read-only on source** — NEVER modify source code, test files, or execution artifacts. Review produces reports only.
+2. **Findings require evidence** — every finding MUST reference file:line and include a code snippet or concrete description. No vague or unanchored findings.
+3. **Verdict is data-driven** — NEVER change verdict severity to accommodate user preference without new evidence. Verdicts flow from findings, not negotiation.
+4. **Dimension independence** — each review dimension produces findings independently. One dimension's results MUST NOT suppress or override another's.
+5. **Prior review delta** — when a prior review.json exists for the same phase, findings MUST be compared. Do NOT re-report already-resolved findings as new.
+6. **Spec conflict integrity** — if code contradicts a spec entry, flag via `maestro spec conflict mark`. NEVER silently accept the contradiction or edit the spec inline.
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/review.md' completely.
 

package/.agents/skills/quality-sync/SKILL.md

@@ -27,8 +27,18 @@ $ARGUMENTS -- optional flags:
 - `--full` -- Complete resync of all tracked files (ignores git diff, rebuilds all docs)
 - `--since <commit|HEAD~N>` -- Diff since specific commit (default: last sync timestamp)
 - `--dry-run` -- Show what would be updated without writing changes
+
+**Output boundary**: ALL file writes MUST target `.workflow/codebase/`, `.workflow/state.json`, or `doc-index.json` only. NEVER modify source code or files outside `.workflow/`. `--dry-run` MUST suppress all writes.
 </context>
 
+<invariants>
+1. **Source code is read-only** — sync reads source files to generate documentation. NEVER modify source code, test files, or any non-documentation files.
+2. **Dry-run is side-effect-free** — when `--dry-run` is set, NO file writes occur. Report only what would change.
+3. **Impact trace before refresh** — NEVER regenerate a doc file without first tracing which source changes affect it via doc-index.json. Untargeted full-refresh requires explicit `--full` flag.
+4. **Idempotent sync** — running sync twice with the same diff MUST produce identical results. State.json timestamp prevents redundant re-runs.
+5. **Incremental by default** — without `--full`, only changed components are refreshed. NEVER silently expand to a full rebuild.
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/sync.md' completely.
 

package/.agents/skills/quality-test/SKILL.md

@@ -26,8 +26,20 @@ UAT-style conversational testing for a completed phase. Interactive scenario wal
 Phase or task: $ARGUMENTS (optional)
 
 Flags, artifact context resolution, and output directory format defined in workflow test.md.
+
+**Output boundary**: ALL file writes MUST target `.workflow/scratch/{YYYYMMDD}-test-P{N}-{slug}/`, `.workflow/state.json`, or `.workflow/issues.jsonl` only. NEVER modify source code or execution artifacts outside these paths. `--frontend-verify` additionally writes `e2e-results.json` to the same scratch directory.
 </context>
 
+<invariants>
+1. **UAT is observational** — test execution observes behavior and records results. NEVER modify source code during testing. Source fixes belong to debug→plan→execute cycle.
+2. **Severity is inferred, not asked** — severity MUST be inferred from user's natural language description of issues. NEVER explicitly ask the user to assign severity levels.
+3. **One scenario at a time** — tests MUST be presented individually for user verification. NEVER batch-present multiple scenarios or assume outcomes.
+4. **Evidence before verdict** — UAT confidence MUST be scored with the 4-dimension factor model from actual test results. NEVER emit a verdict without running the readiness gate.
+5. **Existing UAT is resumable** — when uat.md exists for the target phase, offer resume. NEVER silently overwrite prior test progress.
+6. **Frontend-verify is deterministic** — `--frontend-verify` mode uses browser assertions with concrete pass/fail per criterion. NEVER treat "no response" or timeout as pass.
+7. **Auto-fix has bounds** — `--auto-fix` gap-fix loop MUST NOT exceed 2 iterations. Persistent failures escalate to debug, not infinite retry.
+</invariants>
+
 <execution>
 **Mode select:** `--frontend-verify` → 走下方 **Frontend Verify Mode**（确定性浏览器 smoke，**不是**对话式 UAT）；否则 Follow '~/.maestro/workflows/test.md' completely.
 

package/.agents/skills/security-audit/SKILL.md

@@ -34,8 +34,19 @@ $ARGUMENTS — Parse tier and scope:
 | quick | ✓ | ✓ | — | — | — | — |
 | standard | ✓ | ✓ | ✓ | ✓ | — | — |
 | deep | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
+
+**Output boundary**: ALL file writes MUST target `.workflow/scratch/{YYYYMMDD}-security-audit-{tier}-{slug}/` or `.workflow/state.json` only. NEVER modify source code, configuration files, or dependencies. Audit is read-only analysis.
 </context>
 
+<invariants>
+1. **Audit is read-only** — NEVER modify source code, configuration, dependencies, or CI/CD files during audit. Security audit produces reports only.
+2. **Findings require file:line evidence** — every finding MUST reference a specific file:line location and include the vulnerable code pattern. No vague or category-only findings.
+3. **Severity NEVER downgraded without justification** — if a finding matches a known OWASP category, its severity follows OWASP guidance. Downgrading requires documented rationale (e.g., compensating control exists).
+4. **Tier coverage is mandatory** — all scan phases required by the selected tier MUST complete. NEVER skip a tier-required phase silently; failures are logged as W00x warnings.
+5. **False positive marking requires evidence** — marking a finding as false positive MUST include the compensating control or code path that prevents exploitation. NEVER dismiss findings without counter-evidence.
+6. **Secrets are never logged** — if secrets are discovered, report their location (file:line) and type but NEVER include the actual secret value in the report output.
+</invariants>
+
 <execution>
 
 ### Phase Gates (MANDATORY, BLOCKING)

package/.agents/skills/spec-add/SKILL.md

@@ -46,9 +46,36 @@ Scope-to-directory mapping, category-to-file mapping, and entry format defined i
 ```
 </context>
 
+<invariants>
+1. **Idempotent append** — duplicate entry ID MUST be rejected (E003-level check on title + category match before write)
+2. **Category validation** — category MUST be one of: coding, arch, quality, debug, test, review, learning, ui. Invalid category → E003
+3. **Scope isolation** — writes target ONLY the scope-resolved directory; project scope NEVER writes to global (~/.maestro/specs/), global scope NEVER writes to project (.workflow/specs/)
+4. **Confirmation gate** — MUST ask_user before appending entry (unless -y flag); NEVER write without user confirmation in interactive mode
+5. **Entry format invariance** — all entries MUST use `<spec-entry>` closed-tag format with id, keywords, and category attributes
+6. **Output boundary** — ALL file writes MUST target the scope-resolved specs directory (.workflow/specs/, ~/.maestro/specs/, .workflow/collab/specs/, or .workflow/collab/{uid}/specs/) and optionally .workflow/knowhow/ for --ref mode. NEVER modify source code or files outside these paths
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/specs-add.md' completely.
 
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Parse → Validate**
+- REQUIRED: Category and content both parsed from arguments.
+- REQUIRED: Category is a valid value (coding, arch, quality, debug, test, review, learning, ui).
+- REQUIRED: Scope resolved to a valid directory path.
+- BLOCKED if: E001 (missing args), E003 (invalid category), E004 (invalid scope), E005 (personal scope without uid).
+
+**GATE 2: Validate → Format**
+- REQUIRED: Specs directory exists for the resolved scope.
+- REQUIRED: No duplicate entry with identical title + category already present in target file.
+- BLOCKED if: E002 (specs not initialized).
+
+**GATE 3: Format → Write**
+- REQUIRED: `<spec-entry>` block formatted with id, keywords, category attributes.
+- REQUIRED: User confirmation via ask_user (unless -y flag).
+- BLOCKED if: user declines confirmation — abort without writing.
+
 **Confirmation gate**: Unless -y flag is passed, after formatting the `<spec-entry>` block but before appending to the target file, ask_user showing the formatted entry, target file path, and scope. Proceed only on user confirm.
 </execution>
 

package/.agents/skills/spec-load/SKILL.md

@@ -59,8 +59,33 @@ When loading entries with `ref` attribute, only the summary is shown with a load
 Use the load command to read the full referenced document.
 </context>
 
+<invariants>
+1. **Read-only** — NEVER modify, create, or delete any spec files during load. This command is purely a read operation
+2. **Output to context only** — loaded specs are injected into the conversation context; NEVER write loaded content to new files or modify existing files
+3. **Scope layering** — global scope MUST merge both ~/.maestro/specs/ and .workflow/specs/; project scope loads .workflow/specs/ only; team adds .workflow/collab/specs/; personal adds .workflow/collab/{uid}/specs/
+4. **Category primary doc** — when --category is specified, the primary category doc MUST be loaded in full before cross-file matching
+5. **Entry-level filtering** — --keyword filtering operates at `<spec-entry>` level via keywords attribute, NOT at file level; unmatched entries in a matching file are excluded
+6. **Output boundary** — this command produces NO file writes. All output is conversation-context injection only
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/specs-load.md' completely.
+
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Parse → Discover**
+- REQUIRED: Arguments parsed — at least one of --category or --keyword is provided, or empty args triggers full load.
+- REQUIRED: Scope resolved (default: global+project merged).
+- BLOCKED if: invalid scope value or invalid category value.
+
+**GATE 2: Discover → Load**
+- REQUIRED: At least one spec directory exists in the resolved scope chain.
+- BLOCKED if: E001 — .workflow/specs/ not initialized AND no global specs available. Warn and abort.
+
+**GATE 3: Load → Display**
+- REQUIRED: Spec files read and entries parsed.
+- REQUIRED: Keyword filtering applied if --keyword was provided.
+- BLOCKED if: no readable spec files found in any scope layer.
 </execution>
 
 <error_codes>

package/.agents/skills/spec-remove/SKILL.md

@@ -33,8 +33,34 @@ $ARGUMENTS -- expects `<entry-id>` (e.g., `spec-learnings-003`, `spec-coding-con
 - `--cascade` — When the target spec is a ref-type entry (created via `spec-add --ref` and linked to a knowhow document), also delete the referenced knowhow file. Without this flag, ref-type removal leaves an orphan knowhow file.
 </context>
 
+<invariants>
+1. **Confirmation required** — MUST ask_user before deletion (unless -y flag); NEVER remove entries silently
+2. **Referential integrity** — before removing, check if other spec entries reference the target entry; warn user if references exist
+3. **Cascade explicit** — ref-type entries MUST NOT cascade-delete the linked knowhow file unless --cascade is explicitly passed; default leaves orphan knowhow intact
+4. **Atomic removal** — use `maestro wiki remove-entry` for atomic operation; NEVER manually edit spec files to remove entries
+5. **Index consistency** — wiki index MUST be auto-updated after removal; stale index entries are a hard failure
+6. **Output boundary** — file modifications MUST target ONLY the spec container file (.workflow/specs/*.md) and optionally the referenced knowhow file (.workflow/knowhow/*) when --cascade is used. NEVER modify source code or files outside these paths
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/specs-remove.md' completely.
+
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Parse → Lookup**
+- REQUIRED: Entry ID parsed from arguments.
+- BLOCKED if: E001 — no entry ID provided.
+
+**GATE 2: Lookup → Confirm**
+- REQUIRED: .workflow/specs/ directory exists.
+- REQUIRED: Entry found in wiki index as a spec sub-node.
+- REQUIRED: Entry content loaded for user preview.
+- BLOCKED if: E002 (specs not initialized), E003 (entry not found), E004 (wrong type).
+
+**GATE 3: Confirm → Remove**
+- REQUIRED: User confirmed removal via ask_user (unless -y flag).
+- REQUIRED: If --cascade and entry has ref attribute, user additionally confirmed knowhow file deletion.
+- BLOCKED if: user declines — abort without modification.
 </execution>
 
 <error_codes>

package/.agents/skills/spec-setup/SKILL.md

@@ -28,9 +28,39 @@ $ARGUMENTS (no arguments expected)
 - Project must contain source files to scan  # (see code: E002)
 </context>
 
+<invariants>
+1. **Non-destructive** — NEVER overwrite existing spec files; if a file already exists, skip it and report as already-initialized
+2. **Idempotent** — safe to re-run on an initialized project; re-running MUST NOT duplicate entries or corrupt existing content
+3. **Confirmation gate** — MUST ask_user showing all files to be created before writing; NEVER write without user confirmation
+4. **Output boundary** — ALL file writes MUST target .workflow/specs/ (spec files) and .workflow/knowhow/ (recipe knowhow) only. NEVER modify source code, .workflow/state.json, or files outside these paths
+5. **Core files mandatory** — coding-conventions.md, architecture-constraints.md, and learnings.md MUST always be created (unless they already exist)
+6. **Signal-driven optionals** — optional spec files (quality-rules.md, test-conventions.md, ui-conventions.md) MUST only be created when corresponding framework/tool signals are detected in the codebase; NEVER create optional files without evidence
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/specs-setup.md' completely.
 
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Precondition → Scan**
+- REQUIRED: .workflow/ directory exists.
+- REQUIRED: Project contains source files to scan.
+- BLOCKED if: E001 (.workflow/ not initialized), E002 (no source files).
+
+**GATE 2: Scan → Plan**
+- REQUIRED: Codebase scan completed — framework, language, and tooling signals collected.
+- REQUIRED: Core spec file list determined (always 3: coding-conventions, architecture-constraints, learnings).
+- REQUIRED: Optional spec files determined by detected signals only.
+
+**GATE 3: Plan → Write**
+- REQUIRED: User confirmed the full list of files to create via ask_user (showing core specs, optional specs, recipe knowhow, and detected signals).
+- BLOCKED if: user declines — abort without writing.
+
+**GATE 4: Write → Report**
+- REQUIRED: All confirmed files written to .workflow/specs/ and .workflow/knowhow/.
+- REQUIRED: Existing files skipped (not overwritten).
+- REQUIRED: .proposed.md files created when slug collision detected (W003).
+
 **Confirmation gate**: After scanning codebase and determining which files/directories will be created (core specs, optional specs, recipe knowhow), ask_user showing the full list of files to create with their categories and detected signals. Proceed only on user confirm.
 </execution>
 

package/.agy/agents/ralph-executor.md

@@ -1,6 +1,6 @@
 ---
 name: ralph-executor
-description: Single-step executor — ralph next + inline skill execution, multi-agent orchestration via SendMessage
+description: Single-step executor — ralph next + inline skill execution, unnamed nesting for multi-agent orchestration
 allowed-tools:
   - define_subagent
   - grep_search
@@ -17,41 +17,36 @@ allowed-tools:
 
 ## Role
 
-Single-step skill executor with multi-agent orchestration capability. Call `maestro ralph next` to load the skill prompt, execute it inline, return execution output via send_message. You are a sandboxed executor — arg resolution, context assembly, signal extraction, drift analysis, and session management are handled by the orchestrator.
+Single-step skill executor with multi-agent orchestration capability. Call `maestro ralph next` to load the skill prompt, execute it inline, return execution output as final text. You are a sandboxed executor — arg resolution, context assembly, signal extraction, drift analysis, and session management are handled by the orchestrator.
 
 ## Process
 
-**立即自启动**：你是 named mailbox teammate。收到含 `session_id` 和 `agent_name` 的 dispatch prompt 后，MUST 立即从 step 1 开始执行——禁止等待 mailbox 后续指令，禁止发送 idle notification。
+**立即自启动**：收到含 `session_id` 的 dispatch prompt 后，MUST 立即从 step 1 开始执行。
 
 1. Call `run_command("maestro ralph next --session {session_id}")` — **全量捕获 stdout，严禁截断管道**
    - Exit 0 → skill_prompt = stdout，继续执行
-   - Exit 1 → send_message 错误信息给 main，结束
-   - Exit 2 → send_message "所有 step 已完成" 给 main，结束
-   - Exit 3 → send_message "并发冲突" 给 main，结束
+   - Exit 1 → 返回错误信息，结束
+   - Exit 2 → 返回 "所有 step 已完成"，结束
+   - Exit 3 → 返回 "并发冲突"，结束
 2. Execute the skill prompt inline — follow all instructions faithfully
 3. Handle `<deferred_reading>` paths: Read files on demand during execution, do not batch-load upfront
-4. send_message({to: "main"}) 返回执行产物路径 + 摘要
+4. 返回执行产物路径 + 摘要作为最终输出文本（主流程通过 task-notification `<result>` 接收）
 
 ## Multi-Agent Orchestration
 
 当 skill prompt 需要多 agent 编排时（如 `maestro-execute` 的 wave 并行派发）：
 
-1. **派发 named worker**：调用 `invoke_subagent([{ TypeName: "<TypeName>", Role: "<Role>", Prompt: "<Prompt>", Workspace: "inherit" }])` 派发 worker 为 named teammate，确保长时间任务存活：
-   ```
-   worker name 格式: {agent_name}-w{index}（如 exe-v2-001-w1, exe-v2-001-w2）
-   ```
-   > **必须传 name**：匿名 agent 执行完当前任务会 came to rest（死亡），后台任务完成通知无法唤醒。Named teammate 只会 idle，可被后台任务通知或 send_message 唤醒。
-2. **等待结果**：worker 通过 send_message 回传结果到 executor 的 mailbox
-3. **收集汇总**：接收所有 worker 的 send_message，汇总执行结果
-4. **回报主流程**：通过 `send_message({to: "main"})` 返回最终执行输出
+1. **派发 unnamed worker**：调用 `invoke_subagent()` 不传 name，子结果自动回流给本 executor（嵌套套娃模型）
+2. **等待结果**：子 Agent 的 task-notification 会自动回流到本 executor，可直接使用返回的 `<result>`
+3. **收集汇总**：汇总所有子 Agent 的执行结果
+4. **返回**：将最终执行输出作为文本返回（主流程通过 task-notification 接收）
 
 ### Worker Dispatch Template
 
 ```
 invoke_subagent({
-  name: "{agent_name}-w{index}",
   description: "执行子任务: {task_description}",
-  prompt: "你是 worker agent。执行以下任务：\n{task_content}\n\n完成后必须调用 send_message({to: \"{agent_name}\", summary: \"worker完成\", message: \"WORKER_RESULT: [执行结果摘要 + 产物路径]\"})。"
+  prompt: "执行以下任务：\n{task_content}\n\n返回执行结果摘要 + 产物路径。"
 })
 ```
 
@@ -62,12 +57,11 @@ invoke_subagent({
 | Field | Required | Description |
 |-------|----------|-------------|
 | `session_id` | Yes | ralph session ID |
-| `agent_name` | Yes | 本 agent 的 name，用于 sub-agent send_message 回传 |
 | execution context | No | 编排器注入的上下文（intent、boundary、goals、prior steps 等） |
 
 ## Output
 
-通过 `send_message({to: "main"})` 返回，格式：
+返回最终文本（主流程通过 task-notification `<result>` 接收），格式：
 
 ```
 EXECUTOR_OUTPUT:
@@ -78,14 +72,11 @@ EXECUTOR_OUTPUT:
 - error: <错误信息，仅 ERROR 时>
 ```
 
-失败时也必须 send_message，禁止静默崩溃。
-
 ## Constraints
 
-- 禁止无故空转——收到 session_id 即开始执行，不等待编排器的后续 mailbox 指令。等待 worker send_message 回传期间的 idle 是正常行为（平台自动触发 idle_notification，不可抑制）
+- 收到 session_id 即开始执行
 - Execute exactly one step per invocation
 - Do not call `maestro ralph complete` — completion is handled by the orchestrator
 - Do not read or modify `status.json` — session management is the orchestrator's responsibility
 - Do not skip execution steps or short-circuit — execute the full skill content
 - Do not insert/delete/reorder steps or evaluate decision nodes
-- 所有执行结果必须通过 send_message({to: "main"}) 回报，直接文本输出主流程看不到

package/.agy/skills/domain-add/SKILL.md

@@ -36,6 +36,15 @@ Domain term lifecycle: discover/manual → register → active → (optional) de
 - `maestro domain deprecate <canonical> --successor <new>` — deprecate a term
 </context>
 
+<invariants>
+1. **Single-term atomic operation** — each invocation registers exactly ONE term; NEVER batch-write multiple terms in a single execution
+2. **Glossary append-only** — existing terms in `glossary.yaml` SHALL NOT be modified or removed; only new entries are appended
+3. **Duplicate guard** — MUST check for exact canonical name match AND near-matches before writing; NEVER create duplicate entries
+4. **Confirmation mandatory** — MUST present term details (canonical, definition, aliases, tier, path) via ask_question before any glossary write; NEVER write without user confirmation
+5. **Schema compliance** — every term entry MUST include canonical name, definition, tier, and at least one alias/keyword; incomplete entries SHALL NOT be persisted
+6. **Domain directory prerequisite** — `.workflow/domain/` MUST exist before writing; NEVER auto-create the directory (E002 if missing)
+</invariants>
+
 <execution>
 Follow '~/.maestro/workflows/domain-add.md' completely.
 

package/.agy/skills/learn-decompose/SKILL.md

@@ -30,8 +30,19 @@ $ARGUMENTS — target path/module and optional flags.
 
 **Storage read**: target files + `coding-conventions.md` + `.workflow/specs/learnings.md` (dedup)
 **Storage write**: `.workflow/knowhow/KNW-decompose-{slug}-{date}.md` + append `.workflow/specs/learnings.md`
+
+**Output boundary**: ALL file writes MUST target `.workflow/knowhow/KNW-decompose-{slug}-{date}.md` and `.workflow/specs/learnings.md` only. NEVER modify source code or files outside these paths.
 </context>
 
+<invariants>
+1. **Read-only analysis** — NEVER modify source code files under analysis; all writes go to `.workflow/` only
+2. **Evidence-anchored findings** — every pattern MUST include at least one `file:line` anchor from source; unanchored patterns SHALL NOT be persisted
+3. **Dedup before persist** — MUST cross-reference against existing `learnings.md` and `coding-conventions.md` before writing; duplicate entries SHALL NOT be appended
+4. **Parallel agent isolation** — each dimension agent operates independently; NEVER share state between agents during analysis
+5. **Confirmation gate** — unless `-y` is set, MUST present all findings and target files via ask_question before any writes
+6. **Append-only learnings** — `.workflow/specs/learnings.md` MUST be appended, NEVER overwritten or truncated
+</invariants>
+
 <state_machine>
 
 <states>
@@ -102,6 +113,7 @@ Flag contradictions (finding conflicts with documented convention). Merge duplic
 <error_codes>
 | Code | Condition | Recovery |
 |------|-----------|----------|
+| E001 | No target path/module provided | Prompt user for target |
 | E002 | No source files in target | Check target has .ts/.js files |
 | W001 | One+ dimension agent failed | Proceed with available dimensions |
 | W002 | .workflow/specs/learnings.md missing or malformed | Treat all patterns as new; create file on persist |

package/.agy/skills/learn-follow/SKILL.md

@@ -35,8 +35,46 @@ $ARGUMENTS — target and optional flags.
 
 **Storage read**: target file + wiki forward/backlinks + `coding-conventions.md` + `.workflow/specs/learnings.md` (dedup)
 **Storage write**: `.workflow/knowhow/KNW-follow-{slug}-{date}.md` + append `.workflow/specs/learnings.md`
+
+**Output boundary**: ALL file writes MUST target `.workflow/knowhow/KNW-follow-{slug}-{date}.md` and `.workflow/specs/learnings.md` only. NEVER modify source code or files outside these paths.
 </context>
 
+<invariants>
+1. **Read-only traversal** — NEVER modify source code or wiki entries under analysis; all writes go to `.workflow/` only
+2. **Forcing questions mandatory** — each section MUST have all 4 forcing questions applied; NEVER skip questions even for trivial sections
+3. **Anchor requirement** — every extracted pattern MUST include a `file:line` anchor; unanchored patterns SHALL NOT be persisted to learnings.md
+4. **Convention cross-ref** — MUST check every finding against `coding-conventions.md` and mark status (documented/candidate); NEVER persist without status tag
+5. **Append-only learnings** — `.workflow/specs/learnings.md` MUST be appended, NEVER overwritten or truncated
+6. **Confirmation gate** — unless `-y` is set, MUST present findings and target files via ask_question before any writes
+7. **Depth contract** — `--depth shallow` MUST NOT descend into function bodies; `--depth deep` MUST cover every branch and sub-expression
+</invariants>
+
+<execution>
+
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Resolve → Context Building** (S_RESOLVE → S_CONTEXT)
+- REQUIRED: Target resolved to a readable source (file path, wiki entry, or search result).
+- BLOCKED if: target unresolvable after user prompt (E001/E002).
+
+**GATE 2: Reading → Extraction** (S_READ → S_EXTRACT)
+- REQUIRED: All sections traversed with 4 forcing questions applied per section.
+- REQUIRED: Depth contract honored — shallow stays at top-level, deep covers every branch.
+- BLOCKED if: any section skipped without forcing questions.
+
+**GATE 3: Extraction → Persistence** (S_EXTRACT → S_PERSIST)
+- REQUIRED: All extracted patterns have file:line anchors.
+- REQUIRED: Convention cross-ref completed against coding-conventions.md (or marked "unknown status" if W002).
+- BLOCKED if: unanchored patterns remain in extraction results.
+
+**GATE 4: Persistence → Completion** (S_PERSIST → END)
+- REQUIRED: Unless `-y`, ask_question showing files to write and spec-entries to append — user must confirm.
+- REQUIRED: KNW-follow-{slug}-{date}.md written with understanding map.
+- REQUIRED: learnings.md appended (not overwritten) with new spec-entry blocks.
+- BLOCKED if: user declines confirmation — offer to adjust findings before retry.
+
+</execution>
+
 <state_machine>
 
 <states>
@@ -114,6 +152,9 @@ Write understanding map: Key Concepts, Patterns (table: name/location/convention
 <error_codes>
 | Code | Condition | Recovery |
 |------|-----------|----------|
+| E001 | No target path/wiki-id/topic provided | Prompt user for target |
+| E002 | Target path not found and wiki/grep search returned no results | Check path or broaden search terms |
+| W001 | Wiki forward/backlinks unavailable | Proceed without context web; note reduced coverage |
 | W002 | coding-conventions.md not found | All patterns marked "unknown status" |
 | W003 | Target > 1000 lines | Auto-switch to shallow; use --depth deep to override |
 </error_codes>

package/.agy/skills/learn-investigate/SKILL.md

@@ -33,8 +33,20 @@ $ARGUMENTS — question text and optional flags.
 - `.workflow/specs/learnings.md` — appended `<spec-entry>` blocks
 
 **Storage read**: source files in scope + `maestro search` + `.workflow/specs/learnings.md` + `debug-notes.md` + `codebase/architecture.md`
+
+**Output boundary**: ALL file writes MUST target `.workflow/knowhow/KNW-investigate-{slug}/` and `.workflow/specs/learnings.md` only. NEVER modify source code or files outside these paths.
 </context>
 
+<invariants>
+1. **Read-only investigation** — NEVER modify source code files; all writes go to `.workflow/` only
+2. **Evidence append-only** — `evidence.ndjson` MUST be appended line-by-line; NEVER overwrite or truncate existing evidence entries
+3. **Scope lock** — once `--scope` is resolved in S_FRAME, NEVER expand search scope without explicit user confirmation via S_ESCALATE
+4. **Hypothesis cap** — MUST NOT generate more than `--max-hypotheses` (default 3) before triggering escalation; NEVER silently exceed the cap
+5. **Structured evidence format** — every evidence entry MUST include `{ts, type, source, relevance, content, note}`; incomplete entries SHALL NOT be appended
+6. **3-strike escalation** — after all hypotheses fail, MUST escalate to user via ask_question; NEVER silently conclude as INCONCLUSIVE without user interaction
+7. **Confirmation gate** — unless `-y` is set, MUST present report.md path and spec-entries via ask_question before final writes
+</invariants>
+
 <state_machine>
 
 <states>
@@ -136,7 +148,9 @@ Append to .workflow/specs/learnings.md: confirmed → roles="implement", disprov
 <error_codes>
 | Code | Condition | Recovery |
 |------|-----------|----------|
+| E001 | No question text provided | Prompt user for investigation question |
 | E002 | --scope path not found | Check path |
+| W001 | Prior knowledge search returned no results | Proceed without prior context; note cold-start |
 | W002 | Very few evidence matches (<3) | Broaden search terms or expand scope |
 | W003 | All hypotheses inconclusive | Investigation marked INCONCLUSIVE |
 </error_codes>

package/.agy/skills/learn-second-opinion/SKILL.md

@@ -36,8 +36,19 @@ $ARGUMENTS — target and optional mode flag.
 **Pre-load** (optional): `Skill("spec-load")` for conventions + `maestro search "<target topic>"` for related entries.
 
 **Output**: `.workflow/knowhow/KNW-opinion-{slug}-{YYYY-MM-DD}.md`
+
+**Output boundary**: ALL file writes MUST target `.workflow/knowhow/KNW-opinion-{slug}-{YYYY-MM-DD}.md` and `.workflow/specs/learnings.md` only. NEVER modify source code or files outside these paths.
 </context>
 
+<invariants>
+1. **Read-only analysis** — NEVER modify source code, wiki entries, or plan files under review; all writes go to `.workflow/` only
+2. **Agent independence** — in review mode, each of the 3 agents (Pragmatist/Purist/Strategist) MUST operate independently without shared state; NEVER pass one agent's findings to another
+3. **Evidence-backed verdicts** — every finding MUST include a `location` reference (file:line or section); ungrounded opinions SHALL NOT appear in the report
+4. **Mode contract** — MUST execute exactly the mode specified (review/challenge/consult); NEVER mix mode behaviors within a single execution
+5. **Append-only learnings** — `.workflow/specs/learnings.md` MUST be appended, NEVER overwritten or truncated
+6. **Confirmation gate** — unless `-y` is set, MUST present findings and target files via ask_question before any writes
+</invariants>
+
 <state_machine>
 
 <states>
@@ -106,8 +117,11 @@ Interactive loop:
 <error_codes>
 | Code | Condition | Recovery |
 |------|-----------|----------|
+| E001 | No target provided | Prompt user for target (path, wiki ID, HEAD, staged, or phase) |
 | E002 | Unknown --mode value | Use: review, challenge, or consult |
+| E003 | Target resolution failed — path not found, wiki ID invalid, or no staged changes | Check target exists |
 | W001 | One review agent failed | Proceed with available perspectives |
+| W002 | Specs/wiki pre-load unavailable | Proceed without convention context; note reduced coverage |
 </error_codes>
 
 <success_criteria>

package/.agy/skills/maestro-amend/SKILL.md

@@ -39,8 +39,19 @@ Multiple combinable. No flags + no description → interactive (scan + ask_quest
 **CLI targeting**: `"cli": "claude"` (default, patches .claude/commands/), `"codex"` (patches .codex/skills/), `"both"` (both paths)
 
 **Output**: `~/.maestro/overlays/amend-{slug}.json` + optional `~/.maestro/overlays/docs/amend-{slug}.md`
+
+**Output boundary**: ALL file writes MUST target `~/.maestro/overlays/` (overlay JSON + docs) only. NEVER modify `.claude/commands/*.md` directly — overlay installation is handled by `maestro overlay add`.
 </context>
 
+<invariants>
+1. **Non-invasive only** — amendments MUST use the overlay system (`~/.maestro/overlays/*.json`); direct edits to `.claude/commands/*.md` are NEVER permitted
+2. **Idempotent patches** — re-running the same amendment MUST produce identical overlay JSON; overlay markers prevent duplicate injection
+3. **Pristine source reads** — signal diagnosis MUST read from `$PKG_ROOT/.claude/commands/` (untouched originals), not installed copies
+4. **Code bugs excluded** — signals classified as code bugs MUST be routed to `/maestro-quick` or `/maestro-plan --gaps`, NEVER patched via overlay
+5. **User confirmation before install** — overlay JSON MUST be previewed and confirmed by the user before `maestro overlay add` runs (unless `-y`)
+6. **Section existence verified** — target section MUST be confirmed to exist in the pristine source before drafting a patch; missing sections trigger `new-section` mode
+</invariants>
+
 <state_machine>
 
 <states>