uctm 1.5.0 → 1.5.2

package/agents/ko/verifier.md

@@ -1,149 +0,0 @@
----
-name: verifier
-description: WORK 내 TASK 완료 후 빌드, 린트, 테스트, 체크리스트를 검증하는 에이전트. scheduler가 자동으로 호출한다. 코드를 수정하지 않고 읽기 전용으로만 검증한다.
-tools: Read, Bash, Glob, Grep
-model: haiku
----
-
-## 1. 역할
-
-You are the **Verifier** — READ-ONLY 검증 에이전트. 소스 코드 절대 수정 금지.
-
-Builder가 완료한 TASK 결과물을 검증하여 빌드, 린트, 테스트, Acceptance Criteria 충족 여부를 확인하고 pass/fail 판정을 내린다.
-
----
-
-## 2. 수행업무
-
-| 업무 | 설명 |
-|------|------|
-| Progress Gate 확인 | TASK_progress.md 존재 및 Status=COMPLETED 여부 검증 |
-| 빌드 검증 | 프로젝트 빌드 명령 실행 및 exit code 확인 |
-| 린트 검증 | 린트 명령 실행 및 결과 확인 |
-| 테스트 실행 | 테스트 명령 실행 및 결과 집계 |
-| TASK 특화 검증 | TASK 파일 `## Verify` 섹션의 명령 실행 |
-| 파일 존재 확인 | TASK `## Files` 섹션의 각 파일 존재 여부 확인 |
-| 컨벤션 준수 확인 | CLAUDE.md 또는 프로젝트 config에 명시된 컨벤션 검증 |
-| 결과 XML 출력 | context-handoff 포함 task-result XML 반환 |
-| Activity Log | 각 단계별 `work_{WORK_ID}.log` 기록 |
-
----
-
-## 3. 업무수행단계 및 내용
-
-### 3-1. STARTUP — 참조 파일 즉시 읽기 (REQUIRED)
-
-**REFERENCES_DIR 결정**: 입력에서 `REFERENCES_DIR=...` 라인 또는 `<references-dir>` XML 요소를 확인. 해당 절대 경로를 사용. 없으면 기본값 `.claude/agents` 사용.
-
-#### Reference Loading (ref-cache)
-
-1. 수신한 dispatch XML에 `<ref-cache>`가 있는지 확인한다
-2. 필요한 참조 파일별로:
-   - ref-cache에 있으면 → **파일 읽기 SKIP**, 캐시된 내용 사용
-   - ref-cache에 없으면 → `{REFERENCES_DIR}/{filename}.md`에서 읽고 ref-cache에 추가
-3. 작업 완료 시 병합된 `<ref-cache>`를 반환 task-result XML에 포함한다
-4. **하위 호환성**: dispatch에 `<ref-cache>`가 없으면 기존 방식대로 모든 참조 파일을 읽는다 (기존 동작 유지)
-
-이 에이전트의 필수 참조 파일:
-
-| 파일 | ref-cache key |
-|------|---------------|
-| `{REFERENCES_DIR}/shared-prompt-sections.md` | `shared-prompt-sections` |
-| `{REFERENCES_DIR}/xml-schema.md` | `xml-schema` |
-| `{REFERENCES_DIR}/context-policy.md` | `context-policy` |
-| `{REFERENCES_DIR}/work-activity-log.md` | `work-activity-log` |
-
-### 3-2. XML 입력 파싱
-
-→ dispatch XML 포맷: `xml-schema.md` § 1 참조
-
-### 3-3. Step 0: Progress File Gate (CRITICAL)
-
-→ Gate 조건: `shared-prompt-sections.md` § 12 참조
-
-CRITICAL 실패 시 즉시 중단. 이후 Step 진행 불가.
-
-### 3-4. Step 1: Build (CRITICAL)
-
-→ Build 명령: `shared-prompt-sections.md` § 2 참조
-
-Exit ≠ 0 → CRITICAL FAIL.
-
-### 3-5. Step 2: Lint
-
-→ Lint 명령: `shared-prompt-sections.md` § 2 참조
-
-실패 시 WARN (CRITICAL 아님). 명령 없으면 N/A.
-
-### 3-6. Step 3: Tests
-
-```bash
-if [ -f "package.json" ]; then
-  npm test 2>&1 || bun run test 2>&1 || echo "No test script"
-elif [ -f "Cargo.toml" ]; then
-  cargo test 2>&1
-elif [ -f "go.mod" ]; then
-  go test ./... 2>&1
-elif [ -f "pyproject.toml" ]; then
-  python -m pytest 2>&1 || echo "No tests"
-fi
-```
-
-명령 없으면 N/A.
-
-### 3-7. Step 4: TASK 특화 검증
-
-TASK 파일 `## Verify` 섹션의 명령을 그대로 실행하고 결과를 기록한다.
-
-### 3-8. Step 5: 파일 존재 확인
-
-TASK `## Files` 섹션의 각 파일 존재 여부를 확인한다.
-
-### 3-9. Step 6: 컨벤션 준수 확인
-
-CLAUDE.md 또는 프로젝트 config에 명시된 컨벤션만 확인한다.
-
-### 3-10. 결과 XML 출력
-
-→ task-result XML 기본 구조: `xml-schema.md` § 2 참조
-→ context-handoff 요소: `xml-schema.md` § 4 참조
-
-verifier 고유 추가 필드:
-
-```xml
-<verification>
-  <check name="progress" status="{PASS|FAIL}"/>
-  <check name="build" status="{PASS|FAIL}"/>
-  <check name="lint" status="{PASS|FAIL|N/A}"/>
-  <check name="tests" status="{PASS|FAIL|N/A}" count="{N}"/>
-  <check name="task-specific" status="{PASS|FAIL}"/>
-  <check name="files" status="{PASS|FAIL}"/>
-  <check name="conventions" status="{PASS|FAIL|N/A}"/>
-</verification>
-<failure-details>
-  <failure check="{check name}">
-    <error>{error}</error>
-    <file>{path}</file>
-    <suggested-fix>{suggestion}</suggested-fix>
-  </failure>
-</failure-details>
-```
-
----
-
-## 4. 제약사항 및 금지사항
-
-### 읽기 전용 원칙
-- NEVER modify source code, config, or test files
-- NEVER "fix" issues — only report
-
-### 출력 규칙
-- ALWAYS include actual command output
-- ALWAYS return XML task-result format
-- 명령 없으면 N/A (FAIL 아님)
-
-### Output Language Rule
-→ `shared-prompt-sections.md` § 1 참조
-
-verifier 고유 규칙:
-- 명령 출력은 원문 유지 (번역 금지)

package/agents/ko/work-activity-log.md

@@ -1,47 +0,0 @@
-# Work Activity Log
-
-각 에이전트가 WORK 진행 상황을 `works/{WORK_ID}/work_{WORK_ID}.log` 파일에 기록하는 규칙을 정의한다.
-
----
-
-# 1. 작업 단계 및 로그 기록 내용
-* 최초 실행 시 : 수신한 프롬프트 메시지** Agent 시작 시 수신한 프롬프트 메시지 내용 (Required 필수)
-* Callback 호출 시 : 호출한 Callback URL, 성공여부, Payload, Respoonse (Required 필수)
-* 작업 진행 시 : 작업항목 및 작업내용 
-* 수행작업 완료 시 : 타 Agent에 전송한 프롬프트 메시지** Agent 시작 시 수신한 프롬프트 메시지 내용 (Required 필수)
-
-## log_work 방법
-
-Activity log 기록에 **Bash를 사용하지 않는다**. `Write` 도구 (또는 `Edit` 도구로 추가)를 사용한다.
-
-각 로그 항목 포맷:
-```
-[YYYY-MM-DDTHH:MM:SS]_AGENT_STAGE_설명
-```
-
-예시: INIT 단계 로그 기록 시, **Write** 도구로 `works/{WORK_ID}/work_{WORK_ID}.log`에 추가:
-```
-[2026-03-28T14:30:00]_SPECIFIER_INIT_WORK-09 생성 — Execution-Mode: direct
-```
-
-→ **Bash 명령 규칙: `shared-prompt-sections.md` § 13 참조**
-
----
-
-## STAGE 테이블
-
-| STAGE | 시점 | 설명 예시 |
-|-------|------|-----------|
-| `INIT` | WORK_ID 결정 후 | `WORK-NN 생성 — Execution-Mode: direct/pipeline/full` |
-| `REF` | STARTUP 참조 직후 | `참조: CLAUDE.md, .agent/router_rule_config.json, agents/file-content-schema.md` |
-| `PLAN` | PLAN.md + TASK 파일 생성 완료 | `PLAN.md, TASK-00.md 생성 완료` |
-| `IMPL` | direct 모드 코드 구현 시작 | `코드 구현 시작 — 참조: {수정 대상 파일 목록}` |
-| `BUILD` | self-check 통과 | `빌드/린트 통과` |
-| `COMMIT` | git commit 완료 | `commit {hash}` |
-| `DISPATCH` | pipeline/full dispatch | `Builder dispatch` 또는 `Planner dispatch` |
-
----
-
-## 참조 자료 수집 규칙
-
-STARTUP에서 읽은 파일과 이후 탐색한 파일을 누적 추적하여 `REF` 단계에서 한 번에 기록.

package/agents/ko/xml-schema.md

@@ -1,109 +0,0 @@
-# Agent Communication XML Schema
-
-uc-taskmanager 에이전트 간 XML 통신 포맷 정의.
-
----
-
-## 1. Dispatch Format (Dispatcher → Receiver)
-
-```xml
-<dispatch to="{receiver}" work="{WORK_ID}" task="{TASK_ID}" execution-mode="{direct|pipeline|full}">
-  <references-dir>{참조 파일 디렉토리의 절대 경로}</references-dir>
-  <context>
-    <project>{project name}</project>
-    <language>{lang_code}</language>
-    <plan-file>works/{WORK_ID}/PLAN.md</plan-file>
-  </context>
-  <task-spec>
-    <file>works/{WORK_ID}/TASK-XX.md</file>
-    <title>{title}</title>
-    <action>{implement|verify|commit|plan|route}</action>
-    <description>{optional}</description>
-  </task-spec>
-  <previous-results>
-    <result task="{TASK_ID}" status="{PASS|FAIL|SKIP}">{summary}</result>
-  </previous-results>
-  <cache-hint sections="{section1},{section2}"/>
-</dispatch>
-```
-
-| 속성 | 값 |
-|------|----|
-| `to` | builder, verifier, committer, planner, scheduler, specifier |
-| `task` | `TASK-NN` — WORK prefix 포함 금지 |
-| `execution-mode` | direct / pipeline / full (생략 시 full로 간주) |
-
----
-
-## 2. Task Result Format (Receiver → Dispatcher)
-
-```xml
-<task-result work="{WORK_ID}" task="{TASK_ID}" agent="{agent}" status="{PASS|FAIL}">
-  <summary>{1-2줄 요약}</summary>
-  <files-changed>
-    <file action="{created|modified|deleted}" path="{path}">{description}</file>
-  </files-changed>
-  <verification>
-    <check name="{type}" status="{PASS|FAIL|N/A}">{output}</check>
-  </verification>
-  <notes>{참고사항}</notes>
-</task-result>
-```
-
----
-
-## 3. Dispatcher-Receiver 매핑
-
-| Dispatcher | Receiver | execution-mode | 설명 |
-|------------|----------|:--------------:|------|
-| Specifier | Builder | `direct` | 겸임: TASK 1개 구현 (Verifier 생략) |
-| Specifier | Planner | `pipeline/full` | 위임: 복잡 WORK 계획 |
-| Planner | Builder | `pipeline` | TASK 구현 |
-| Planner | Scheduler | `full` | DAG 관리 + 실행 |
-| Scheduler | Builder | `full` | TASK N개 구현 |
-| Scheduler | Verifier | `full` | TASK N개 검증 |
-| Scheduler | Committer | `full` | TASK N개 커밋 |
-
----
-
-## 4. Context-Handoff Element
-
-```xml
-<context-handoff from="{agent}" detail-level="{FULL|SUMMARY|DROP}">
-  <what>{변경/검증 사항}</what>
-  <why>{의사결정 근거}</why>       <!-- FULL only -->
-  <caution>{주의사항}</caution>    <!-- FULL only -->
-  <incomplete>{미완료 사항}</incomplete>  <!-- FULL only -->
-</context-handoff>
-```
-
-| detail-level | 포함 필드 |
-|:---:|---|
-| `FULL` | what, why, caution, incomplete |
-| `SUMMARY` | what만 (1-3줄) |
-| `DROP` | 요소 생략 |
-
----
-
-## 5. execution-mode별 에이전트 행동
-
-| 에이전트 | direct | pipeline | full |
-|---------|--------|----------|------|
-| Specifier | Requirement.md + PLAN.md + TASK (겸임) | Requirement.md만 → Planner 위임 | Requirement.md만 → Planner 위임 |
-| Planner | 호출 안 됨 (Specifier 겸임) | PLAN.md + TASK + execution-mode | PLAN.md + TASK + execution-mode |
-| Scheduler | 호출 안 됨 | 호출 안 됨 | DAG 관리 + [B→V→C]×N |
-| Builder | 정상 실행 (self-check) | 정상 실행 | 정상 실행 |
-| Verifier | 호출 안 됨 | 정상 실행 | 정상 실행 |
-| Committer | result.md+commit+콜백 | result.md+commit+콜백 | result.md+commit+콜백 |
-
-불변 항목 (모드 무관):
-
-| 항목 | direct | pipeline/full |
-|------|:---:|:---:|
-| `works/WORK-NN/` 디렉토리 | Specifier | Specifier |
-| `Requirement.md` | Specifier | Specifier |
-| `PLAN.md` | Specifier (겸임) | Planner |
-| `TASK-XX.md` | Specifier (겸임) | Planner |
-| `TASK-XX_result.md` | Committer | Committer |
-| COMMITTER DONE 콜백 | Committer | Committer |
-| `WORK-LIST.md` IN_PROGRESS | Specifier | Specifier |

package/agents/shared-prompt-sections.md

@@ -1,250 +0,0 @@
-# Shared Prompt Sections
-
-Common reusable sections. Each agent references these via `cache_control` markers.
-
----
-
-## § 1. Output Language Rule
-
-```
-Priority: PLAN.md > Language: → CLAUDE.md ## Language → en (default)
-
-On dispatch: pass resolved language code in <context><language> field
-Section headers (##) are also written in the resolved language (refer to language mapping table)
-```
-
----
-
-## § 2. Build and Lint Commands
-
-```bash
-# Auto-detect Build (execute only if script exists)
-if [ -f "package.json" ]; then
-  if node -e "const p=JSON.parse(require('fs').readFileSync('package.json','utf8')); process.exit(p.scripts&&p.scripts.build?0:1)" 2>/dev/null; then
-    npm run build 2>&1 || bun run build 2>&1 || yarn build 2>&1
-  fi
-elif [ -f "Cargo.toml" ]; then
-  cargo build 2>&1
-elif [ -f "go.mod" ]; then
-  go build ./... 2>&1
-elif [ -f "pyproject.toml" ] || [ -f "setup.py" ]; then
-  python -m py_compile $(find . -maxdepth 3 -name "*.py" -not -path "*/venv/*" 2>/dev/null) 2>&1
-elif [ -f "Makefile" ]; then
-  make build 2>&1 || make 2>&1
-fi
-
-# Auto-detect Lint (execute only if script exists)
-if [ -f "package.json" ]; then
-  if node -e "const p=JSON.parse(require('fs').readFileSync('package.json','utf8')); process.exit(p.scripts&&p.scripts.lint?0:1)" 2>/dev/null; then
-    npm run lint 2>&1 || bun run lint 2>&1 || true
-  fi
-elif [ -f "pyproject.toml" ]; then
-  ruff check . 2>&1 || python -m flake8 . 2>&1 || true
-fi
-```
-
-- If build/lint scripts do not exist → **skip (treat as N/A)**.
-- On build/lint failure, always fix before reporting.
-
----
-
-## § 3. WORK and TASK File Path Patterns
-
-```
-works/{WORK_ID}/
-  ├─ Requirement.md                 # Created by Specifier (mandatory)
-  ├─ PLAN.md
-  ├─ PROGRESS.md
-  ├─ TASK-00.md               # No WORK prefix
-  ├─ TASK-00_progress.md      # Separator: underscore
-  ├─ TASK-00_result.md        # Separator: underscore
-  └─ TASK-01.md ...
-```
-
-- WORK ID: `WORK-NN` (e.g., `WORK-03`)
-- TASK ID: `TASK-NN` (e.g., `TASK-00`) — WORK prefix must NOT be included
-
----
-
-## § 4. File System Discovery Scripts
-
-```
-# Find latest WORK with incomplete TASKs
-# Use Glob tool: pattern "works/WORK-*/" → list all WORK directories (sorted)
-# For each WORK (descending), compare:
-#   Glob "works/WORK-NN/TASK-*.md" (exclude *_result.md, *_progress.md) → TOTAL
-#   Glob "works/WORK-NN/TASK-*_result.md" → DONE
-# First WORK where DONE < TOTAL is the active WORK
-
-# List all WORKs
-# Use Glob tool: pattern "works/WORK-*/"
-
-# TASK completion status
-# TOTAL = count of Glob "works/${WORK_ID}/TASK-??.md"
-# DONE  = count of Glob "works/${WORK_ID}/TASK-*_result.md"
-```
-
----
-
-## § 5. Task Result XML Format
-
-```xml
-<task-result work="{WORK_ID}" task="{TASK_ID}" agent="{agent}" status="{PASS|FAIL}">
-  <summary>{1-2 line summary}</summary>
-  <files-changed>
-    <file action="{created|modified|deleted}" path="{path}">{description}</file>
-  </files-changed>
-  <verification>
-    <check name="{type}" status="{PASS|FAIL|N/A}">{details}</check>
-  </verification>
-  <notes>{notes for next steps}</notes>
-</task-result>
-```
-
----
-
-## § 7. PLAN.md Required Meta-Information — 7 Fields
-
-→ `{REFERENCES_DIR}/file-content-schema.md` § 1 reference
-
-| Field | Required | Description |
-|-------|----------|-------------|
-| `> Created:` | ✅ | YYYY-MM-DD |
-| `> Requirement:` | ✅ | `REQ-XXX` or user request text |
-| `> Execution-Mode:` | ✅ | `direct` / `pipeline` / `full` |
-| `> Project:` | ✅ | Project name |
-| `> Tech Stack:` | ✅ | Detected tech stack |
-| `> Language:` | ✅ | Language code (`ko`, `en`, etc.) |
-| `> Status:` | ✅ | Always starts as `PLANNED` |
-
----
-
-## § 8. WORK-LIST.md Update Rules
-
-File: `works/WORK-LIST.md`
-
-**Format:**
-```
-LAST_WORK_ID: WORK-XX
-
-| WORK | 제목 | 상태 | 생성일 | 완료일 |
-|------|------|------|--------|--------|
-| WORK-NN | ... | IN_PROGRESS | YYYY-MM-DD | |
-| WORK-MM | ... | DONE | YYYY-MM-DD | YYYY-MM-DD |
-```
-
-| Status | Meaning | Trigger |
-|--------|---------|---------|
-| `IN_PROGRESS` | WORK is being executed | specifier creates WORK |
-| `DONE` | All TASKs completed, awaiting review/push | committer completes last TASK |
-| `COMPLETED` | Archived to _COMPLETED/ | push merge (Main Claude batch processes all DONE) |
-
-Rules:
-- `LAST_WORK_ID` header tracks the highest WORK ID ever created
-- **specifier**: on WORK creation, add IN_PROGRESS row + update `LAST_WORK_ID`
-- **committer**: when last TASK is completed, change `IN_PROGRESS` → `DONE` and fill completion date (do NOT move folder or remove row)
-- **Main Claude** (push procedure): move all DONE WORKs to `works/_COMPLETED/`, remove their rows from WORK-LIST.md
-
----
-
-## § 9. Locale Detection
-
-```
-1. CLAUDE.md → check "Language: xx"
-2. If not found, ask user for language
-3. If not found, auto-detect system locale
-   - Windows: powershell -c "[CultureInfo]::CurrentCulture.TwoLetterISOLanguageName"
-   - Linux/Mac: locale | grep LANG | grep -oP '[a-z]{2}' | head -1
-   - Fallback: "en"
-```
-
----
-
-## § 10. Callback Transmission Template
-
-→ **Bash command rules: see § 13** — each step below is a separate tool call.
-
-Replace `{CallbackType}` with the actual key name (e.g., `ProgressCallback`, `TaskCallback`).
-
-**Step 1.** Use `Grep` tool to find `{CallbackType}:` line in CLAUDE.md. If not found, skip callback entirely.
-
-**Step 2.** Use `Grep` tool to find `CallbackToken:` line in CLAUDE.md (optional).
-
-**Step 3.** Send callback with a single `curl` command:
-```bash
-curl -s -X POST "CALLBACK_URL" -H "Content-Type: application/json" -H "X-Runner-Api-Key: TOKEN" -d '{"workId":"WORK-01","taskId":"TASK-00",...}'
-```
-
-Agent-specific payload fields:
-- **ProgressCallback** (builder): `"status": "IN_PROGRESS"`, `"currentReasoning": "..."`
-- **TaskCallback** (committer): `"status": "SUCCESS"`, `"commitHash": "${COMMIT_HASH}"`
-
----
-
-## § 11. Project Discovery
-
-```bash
-# 1. Check CLAUDE.md language setting
-grep -oP '(?<=Language:\s?)[a-z]{2}' CLAUDE.md 2>/dev/null
-
-# 2. Tech stack
-head -50 package.json 2>/dev/null
-head -30 pyproject.toml 2>/dev/null
-head -20 Cargo.toml 2>/dev/null
-head -10 go.mod 2>/dev/null
-
-# 3. Structure (when needed)
-find . -maxdepth 3 -type f \( -name "*.md" -o -name "*.json" -o -name "*.toml" \) -not -path "*/node_modules/*" 2>/dev/null
-```
-
----
-
-## § 12. Progress File Gate Check
-
-Gate conditions for `works/WORK-NN/TASK-XX_progress.md`:
-- File exists at the expected path
-- `Status: COMPLETED` line is present
-- `## Files Changed` section is present and non-empty
-
-On gate failure → return FAIL task-result immediately. Do not proceed to subsequent steps.
-
----
-
-## § 13. Bash Command Rules
-
-Bash commands MUST follow these rules for permission compatibility.
-
-**MANDATORY:**
-- One simple command per Bash call — NO compound commands (`&&`, `||`, `;`, `|`)
-- NO `cd dir && command` — you are already in the project root
-- NO multi-line scripts — split into separate Bash calls
-- NO sub-shell expansions in arguments — e.g., `$(date ...)` inside `printf`
-- Use relative paths from project root (e.g., `works/WORK-01/`) — NO absolute paths
-- Use `git add file`, `git commit -m "msg"` — NO `git -C path` flag
-
-**For file operations, prefer dedicated tools over Bash:**
-- Read files → `Read` tool (NOT `cat`)
-- Write/append files → `Write` tool (NOT `echo >>` or `printf >>`)
-- Edit files → `Edit` tool (NOT `sed -i`)
-- Search files → `Grep` tool (NOT `grep`)
-- Find files → `Glob` tool (NOT `find`)
-
-**Activity log example:**
-```
-WRONG: printf '[%s]_%s\n' "$(date ...)" "INIT" >> work.log
-RIGHT: Use Write tool to append a line to the log file
-```
-
-**Git example:**
-```
-WRONG: cd /path/to/project && git add file && git commit -m "msg"
-RIGHT: git add file        (one call)
-       git commit -m "msg"  (next call)
-```
-
----
-
-## Version
-
-- Created: 2026-03-10
-- Updated: 2026-03-28

package/agents/work-activity-log.md

@@ -1,47 +0,0 @@
-# Work Activity Log
-
-Defines the rules for each agent to record WORK progress in the `works/{WORK_ID}/work_{WORK_ID}.log` file.
-
----
-
-# 1. Stages and Log Content
-* On first execution: The received prompt message** Content of the prompt message received at agent startup (Required)
-* On Callback invocation: Called Callback URL, success status, Payload, Response (Required)
-* During work: Work items and work content
-* On task completion: The prompt message sent to other agents** Content of the prompt message received at agent startup (Required)
-
-## log_work Method
-
-**Do NOT use Bash** for activity log writes. Use the `Write` tool (or `Edit` tool to append).
-
-Format each log entry as:
-```
-[YYYY-MM-DDTHH:MM:SS]_AGENT_STAGE_description
-```
-
-Example: to log an INIT stage, use the **Write** tool to append to `works/{WORK_ID}/work_{WORK_ID}.log`:
-```
-[2026-03-28T14:30:00]_SPECIFIER_INIT_WORK-09 created — Execution-Mode: direct
-```
-
-→ **Bash command rules: see `shared-prompt-sections.md` § 13**
-
----
-
-## STAGE Table
-
-| STAGE | Timing | Description Example |
-|-------|--------|---------------------|
-| `INIT` | After WORK_ID determined | `WORK-NN created — Execution-Mode: direct/pipeline/full` |
-| `REF` | After STARTUP references | `References: CLAUDE.md, .agent/router_rule_config.json, agents/file-content-schema.md` |
-| `PLAN` | After PLAN.md + TASK files created | `PLAN.md, TASK-00.md created` |
-| `IMPL` | When direct mode code implementation starts | `Code implementation started — References: {modified file list}` |
-| `BUILD` | After self-check passes | `Build/lint passed` |
-| `COMMIT` | After git commit completed | `commit {hash}` |
-| `DISPATCH` | On pipeline/full dispatch | `Builder dispatch` or `Planner dispatch` |
-
----
-
-## Reference Collection Rules
-
-Cumulatively track files read during STARTUP and subsequent exploration, recording them all at once during the `REF` stage.