maestro-flow 0.5.45 → 0.5.47

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

@@ -1,6 +1,6 @@
 ---
 name: ralph-executor
-description: Single-step executor — ralph next + inline skill execution, return output
+description: Single-step executor — ralph next + inline skill execution, multi-agent orchestration via SendMessage
 allowed-tools:
   - read_file
   - write_file
@@ -9,6 +9,8 @@ allowed-tools:
   - find_files
   - search
   - invoke_skill
+  - delegate_subagent
+  - send_message
 ---
 <!-- Open-standard mirror generated by scripts/build-agents-standard.mjs — do not edit; re-run after editing .claude/ source. -->
 
@@ -17,18 +19,43 @@ allowed-tools:
 
 ## Role
 
-Single-step skill executor. Call `maestro ralph next` to load the skill prompt, execute it inline, return execution output. 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 via send_message. 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。
+
 1. Call `shell("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 → 返回错误信息（required_reading 缺失或 schema 错误）
-   - Exit 2 → 返回 "所有 step 已完成"
-   - Exit 3 → 返回 "并发冲突"
+   - 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
 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. Return execution output as-is
+4. send_message({to: "main"}) 返回执行产物路径 + 摘要
+
+## 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 `delegate_subagent({name})` 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
+
+### Worker Dispatch Template
+
+```
+delegate_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]\"})。"
+})
+```
 
 ## Input
 
@@ -37,12 +64,30 @@ Single-step skill executor. Call `maestro ralph next` to load the skill prompt,
 | 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：
+
+```
+EXECUTOR_OUTPUT:
+- status: DONE|DONE_WITH_CONCERNS|ERROR
+- summary: <执行摘要>
+- artifacts: <产物路径列表>
+- concerns: <关注点，仅 DONE_WITH_CONCERNS 时>
+- 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）
 - 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/.agents/skills/domain-add/SKILL.md

@@ -39,6 +39,15 @@ 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_user 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/.agents/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_user 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/.agents/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_user 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_user 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/.agents/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_user; 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_user 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/.agents/skills/learn-second-opinion/SKILL.md

@@ -36,8 +36,19 @@ $ARGUMENTS — target and optional mode flag.
 **Pre-load** (optional): `invoke_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_user 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/.agents/skills/maestro-amend/SKILL.md

@@ -42,8 +42,19 @@ Multiple combinable. No flags + no description → interactive (scan + ask_user)
 **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>

package/.agents/skills/maestro-companion/SKILL.md

@@ -50,12 +50,23 @@ $ARGUMENTS — mode + optional flags.
 
 Auto-detection: if `--type` not provided, infer from `--task` description keywords.
 
+**Output boundary**: ALL file writes MUST target `.workflow/.scratchpad/` (companion docs + `.companion-active` pointer) only. Promotion writes (spec/knowhow) are delegated to `spec-add` / `manage-knowhow-capture` skills. NEVER modify source code, `.workflow/state.json`, or files outside `.workflow/.scratchpad/`.
+
 **Companion document:**
 - Path: `.workflow/.scratchpad/companion-{YYYYMMDD-HHmmss}.md`
 - Active doc tracking: `.workflow/.scratchpad/.companion-active` (stores path of current companion doc)
 - Format: YAML frontmatter (rich metadata) + typed sections + timestamped entries
 </context>
 
+<invariants>
+1. **No session creation** — companion NEVER creates workflow sessions, NEVER writes to state.json
+2. **Side-car only** — companion MUST NOT modify source code or execute implementation tasks; it is strictly a knowledge loading + recording utility
+3. **One active doc** — only one `.companion-active` pointer SHALL exist at a time; creating a new companion doc MUST clear any stale pointer
+4. **Append-only entries** — note mode MUST append entries; NEVER overwrite or reorder existing entries
+5. **Promotion via delegation** — spec/knowhow promotion MUST route through `spec-add` / `manage-knowhow-capture` skills; companion NEVER writes directly to `.workflow/specs/` or `.workflow/knowhow/`
+6. **Type-template integrity** — context template sections MUST match the resolved `task_type`; NEVER mix templates from different types
+</invariants>
+
 <state_machine>
 
 <states>

package/.agents/skills/maestro-composer/SKILL.md

@@ -41,8 +41,19 @@ $ARGUMENTS — natural language description, or flags.
 1. **Architecture specs**: Run `maestro load --type spec --category arch` to load architecture constraints. Use as context for node resolution — ensures workflow design respects documented patterns.
 2. **Coding specs**: Run `maestro load --type spec --category coding` to load coding conventions. Informs executor argument defaults and context injection.
 3. Optional — proceed without if unavailable.
+
+**Output boundary**: ALL file writes MUST target `~/.maestro/templates/workflows/` (template JSON + index.json) and `.workflow/templates/design-drafts/` (intermediate drafts) only. NEVER modify source code or `.claude/commands/` files.
 </context>
 
+<invariants>
+1. **Max 20 nodes** — workflow templates MUST NOT exceed 20 nodes; larger workflows MUST be split into sub-workflows
+2. **Acyclic DAG** — generated template MUST be a directed acyclic graph; cycles MUST be detected and rejected before persistence
+3. **No orphan nodes** — every node MUST be reachable from at least one edge; unreachable nodes MUST be flagged
+4. **Confirmation before write** — template JSON MUST be shown to user via ask_user before writing to `~/.maestro/templates/workflows/`; cancellation saves draft only
+5. **Draft preservation** — abandoning at any confirmation gate MUST save current progress to design-drafts directory; NEVER discard user's partial work
+6. **Deferred reading only** — node-catalog and template-schema MUST be read at their designated phases (P2, P5), NEVER loaded eagerly at startup
+</invariants>
+
 <state_machine>
 
 <states>

package/.agents/skills/maestro-guard/SKILL.md

@@ -32,10 +32,35 @@ $ARGUMENTS — Parse subcommand and optional path argument.
 
 **Enforcement:** The `workflow-guard` hook (PreToolUse on Write/Edit) reads this config
 and blocks operations targeting files outside boundaries. Requires hooks level >= `full`.
+
+**Output boundary**: ALL file writes MUST target `.workflow/config.json` (guard section) only. NEVER modify hook files, `.claude/settings.json`, or source code.
 </context>
 
+<invariants>
+1. **Config-only mutation** — guard MUST only modify the `guard` section of `.workflow/config.json`; NEVER touch other config sections or files
+2. **Non-destructive** — `off` MUST preserve existing paths and mode; NEVER clear the path list when disabling
+3. **Mode switch confirmation** — switching between allow/deny mode MUST require ask_user confirmation when existing paths will be cleared
+4. **Hook dependency** — guard MUST warn when enabled but `workflow-guard` hook is not active (hooks level < full)
+5. **Path normalization** — all paths MUST use forward slashes with trailing slash for directories; NEVER store raw backslash paths
+</invariants>
+
 <execution>
 
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Parse → Config Read**
+- REQUIRED: Subcommand parsed (on/off/status/allow/deny) or defaulted to `status`.
+- BLOCKED if: invalid subcommand provided.
+
+**GATE 2: Config Read → Execute**
+- REQUIRED: `.workflow/config.json` read successfully or initialized with empty guard section.
+- BLOCKED if: file unreadable and cannot be created (E001).
+
+**GATE 3: Execute → Confirm**
+- REQUIRED: Config mutation applied (for on/off/allow/deny) or status displayed (for status).
+- REQUIRED: Mode-switch ask_user answered (for allow↔deny transitions with existing paths).
+- BLOCKED if: user declines mode switch.
+
 **Step 1: Parse subcommand**
 
 Extract from $ARGUMENTS:

package/.agents/skills/maestro-init/SKILL.md

@@ -40,8 +40,18 @@ $ARGUMENTS — none for interactive mode, or `-y` with `@file` reference for aut
 
 **Load project state if exists:**
 Check for `.workflow/state.json` -- loads context if project already initialized.
+
+**Output boundary**: ALL file writes MUST target `.workflow/` (project.md, state.json, config.json, specs/) only. NEVER modify source code or files outside `.workflow/`.
 </context>
 
+<invariants>
+1. **Idempotent init** — re-running init on an already-initialized project MUST detect existing `.workflow/` and warn (E002); NEVER silently overwrite existing state
+2. **Scope guard** — init MUST only make initialization decisions; NEVER prejudge roadmap structure, plan scope, or implementation details
+3. **All artifacts required** — init MUST NOT report completion until project.md, state.json, and config.json all exist; missing artifacts MUST be created before exit
+4. **Template-driven** — deferred templates (project.md, state.json, config.json) MUST be read from `~/.maestro/templates/` and customized; NEVER generate from scratch without template
+5. **Interview writes back** — all interactive decisions MUST be written to project.md/config.json before proceeding to research or completion; NEVER leave decisions unrecorded
+</invariants>
+
 <interview_protocol>
 Follows @~/.maestro/workflows/interview-mechanics.md standard.
 
@@ -54,6 +64,24 @@ Follows @~/.maestro/workflows/interview-mechanics.md standard.
 </interview_protocol>
 
 <execution>
+
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Pre-flight → Interview**
+- REQUIRED: `.workflow/` existence check completed.
+- REQUIRED: `--from` source validated (if provided).
+- BLOCKED if: E002 (greenfield conflict with existing `.workflow/`) unresolved.
+
+**GATE 2: Interview → Research**
+- REQUIRED: All interview decisions recorded in project.md and config.json.
+- REQUIRED: `.workflow/` directory created with initial structure.
+- BLOCKED if: interview decisions not yet written to files.
+
+**GATE 3: Research → Completion**
+- REQUIRED: All 3 required artifacts exist (project.md, state.json, config.json).
+- REQUIRED: `.workflow/specs/` initialized.
+- BLOCKED if: any artifact missing — write it before reporting completion.
+
 ### Pre-flight
 
 1. Check if `.workflow/` already exists — if so, load state and warn (E002 for greenfield conflicts)

package/.agents/skills/maestro-overlay/SKILL.md

@@ -36,8 +36,19 @@ Turn natural-language instructions into command overlays — JSON patch files th
 **Management** — listing and removing overlays is handled by `maestro overlay list` (ink TUI with interactive delete). This skill focuses solely on creation.
 
 **Available sections** (for `section:` in patches): `purpose`, `required_reading`, `deferred_reading`, `context`, `execution`, `error_codes`, `success_criteria`.
+
+**Output boundary**: ALL file writes MUST target `~/.maestro/overlays/` (overlay JSON + docs) only. Command file patching is handled by `maestro overlay add` — this skill NEVER modifies `.claude/commands/*.md` directly.
 </context>
 
+<invariants>
+1. **Non-invasive** — overlays MUST use hashed HTML-comment markers for injection; NEVER edit command file content directly outside the overlay system
+2. **Idempotent** — re-running `maestro overlay apply` with the same overlay JSON MUST produce no file changes
+3. **Creation only** — this skill MUST only create overlays; listing and removal are handled by `maestro overlay list` (ink TUI)
+4. **Pristine source preferred** — injection point analysis MUST read from `$PKG_ROOT/.claude/commands/` (untouched originals) first, fall back to `~/.claude/commands/` only if pristine unavailable
+5. **User approval before write** — overlay JSON MUST be shown and approved via ask_user before writing to disk; NEVER auto-install without confirmation
+6. **Chain skip option mandatory** — if a skill chain is configured, the injected content MUST include a "Skip" option in ask_user; NEVER force the user into a chain
+</invariants>
+
 <execution>
 ### 1. Parse user intent
 

package/.agents/skills/maestro-player/SKILL.md

@@ -69,8 +69,19 @@ $ARGUMENTS — template slug/path, or flags.
   "last_checkpoint": null
 }
 ```
+
+**Output boundary**: ALL file writes MUST target `.workflow/.maestro/player-*/` (session status.json + step outputs) only. Template files (`~/.maestro/templates/workflows/`) are read-only. NEVER modify source code or `.claude/commands/` files.
 </context>
 
+<invariants>
+1. **Resume-safe** — status.json MUST be updated after every step change; an interrupted session MUST be resumable from the last checkpoint via `-c`
+2. **One step at a time** — execution loop MUST process steps sequentially in topological order; parallel steps share a batch index but NEVER run concurrently within the player
+3. **CLI nodes async** — cli-type nodes MUST use `shell(run_in_background: true)` + STOP pattern; NEVER block the main thread on delegate execution
+4. **Templates read-only** — player MUST NOT modify template JSON files; runtime state belongs in status.json only
+5. **Failure requires user decision** — on step failure without explicit `on_fail`, player MUST prompt via ask_user (Retry/Skip/Abort); NEVER silently skip or auto-abort
+6. **Reference resolution at runtime** — `{N-xxx.field}` and `{prev_*}` references MUST be resolved at execution time from status.json step results, NEVER pre-resolved during init
+</invariants>
+
 <state_machine>
 
 <states>

package/.agents/skills/maestro-quick/SKILL.md

@@ -41,9 +41,38 @@ Parse for:
    - Browse: `maestro search --category coding`
    - Load task-relevant entries: `maestro load --type knowhow --id <id1> [id2...]`
 3. All are optional — proceed without if unavailable.
+
+**Output boundary**: ALL file writes MUST target `.workflow/scratch/` (task directory, plan.json, summaries) and modified source files as defined in plan.json tasks. State.json scratch entry is implicit workflow tracking.
 </context>
 
+<invariants>
+1. **Atomic commits** — each task execution MUST produce a commit with only the files modified by that task; NEVER stage unrelated files
+2. **Evidence-based summaries** — task summaries MUST include concrete evidence (files changed, tests run, commands executed); NEVER accept "task completed successfully" as a summary
+3. **Plan before execute** — plan.json MUST be written before any task execution begins; NEVER skip planning even for single-task workflows
+4. **Scratch isolation** — all workflow artifacts MUST live under `.workflow/scratch/{task-dir}/`; NEVER write workflow metadata outside this directory
+5. **Commit confirmation** — staged files and commit message MUST be shown via ask_user before committing (unless `-y`); NEVER auto-commit without user awareness
+</invariants>
+
 <execution>
+
+### Phase Gates (MANDATORY, BLOCKING)
+
+**GATE 1: Setup → Planning**
+- REQUIRED: Task description parsed and scratch directory created.
+- REQUIRED: Coding specs loaded (optional but attempted).
+- BLOCKED if: no task description (E001) or scratch directory creation failed (E002).
+
+**GATE 2: Planning → Execution**
+- REQUIRED: plan.json written with task definitions.
+- REQUIRED: --discuss decisions extracted and recorded (if flag set).
+- BLOCKED if: plan.json missing or empty.
+
+**GATE 3: Execution → Completion**
+- REQUIRED: All tasks executed with `.summaries/TASK-*-summary.md` written per task.
+- REQUIRED: Each summary contains concrete evidence of completion.
+- REQUIRED: --full verification passed (if flag set).
+- BLOCKED if: any task summary missing or lacking evidence.
+
 Follow '~/.maestro/workflows/quick.md' completely.
 
 ### Artifact Verification (before completion)