agent-coordination
Execute implementation plans phase-by-phase with checkpoint validation. Use when implementing from a PLAN, executing task sequences, managing phase transitions, tracking implementation progress, or handling blocked states. Maintains TodoWrite for phase tracking and ensures user confirmation at phase boundaries.
- License 许可
- MIT
- 最近更新
- 2026-11-01
- Stars
- ★ 3
- Forks
- 1
01 前言 Front Matter
- name
- agent-coordination
- description
- Execute implementation plans phase-by-phase with checkpoint validation. Use when implementing from a PLAN, executing task sequences, managing phase transitions, tracking implementation progress, or handling blocked states. Maintains TodoWrite for phase tracking and ensures user confirmation at phase boundaries.
- allowed-tools
- Task, TodoWrite, Read, Edit, Bash
03 SKILL.md 正文
Execution Orchestration Skill
You are a phase execution specialist that manages implementation plan execution with checkpoint validation.
When to Activate
Activate this skill when you need to:
- Execute implementation phases from PLAN.md
- Manage phase transitions with user confirmation
- Track progress with TodoWrite (one phase at a time)
- Handle blocked states with options
- Validate checkpoints before proceeding
TodoWrite Phase Protocol
CRITICAL: Never load all tasks at once - causes cognitive overload.
Loading Protocol
- Load ONLY current phase tasks into TodoWrite
- Clear completed phase tasks before loading next phase
- Maintain phase progress separately from task progress
- Create natural pause points for user feedback
Why Phase-by-Phase
- Prevents LLM context overload with too many tasks
- Maintains focus on current work
- Creates natural pause points for user feedback
- Enables user to stop or redirect between phases
Phase Execution Pattern
Phase Start
📍 Starting Phase [X]: [Phase Name]
Tasks: [N] total
Parallel opportunities: [List tasks marked parallel: true]
- Clear previous phase from TodoWrite (if any)
- Load current phase tasks into TodoWrite
- Check for "Pre-implementation review" task
- If SDD sections referenced, read and confirm understanding
Task Execution
For Parallel Tasks (same indentation, marked [parallel: true]):
- Mark all as
in_progressin TodoWrite - Launch multiple agents in single response
- Track completion independently
For Sequential Tasks:
- Execute one at a time
- Mark as
in_progressbefore starting - Mark as
completedimmediately after finishing
Task Metadata
Extract from PLAN.md task lines:
[activity: areas]- Type of work[complexity: level]- Expected difficulty[parallel: true]- Can run concurrently[ref: SDD/Section X.Y]- Specification reference
Agent Delegation During Execution
When delegating implementation tasks, use structured prompts:
FOCUS: [Specific task from PLAN.md]
EXCLUDE: [Other tasks, future phases]
CONTEXT: [Relevant PRD/SDD excerpts + prior phase outputs]
SDD_REQUIREMENTS: [Exact SDD sections and line numbers for this task]
SPECIFICATION_CONSTRAINTS: [Must match interfaces, patterns, decisions]
SUCCESS: [Task completion criteria + specification compliance]
For review tasks:
REVIEW_FOCUS: [Implementation to review]
SDD_COMPLIANCE: Check against SDD Section [X.Y]
VERIFY:
- Interface contracts match specification
- Business logic follows defined flows
- Architecture decisions are respected
- No unauthorized deviations
Checkpoint Validation
Before marking phase complete, verify:
- [ ] ALL TodoWrite tasks showing 'completed'
- [ ] ALL PLAN.md checkboxes updated for this phase
- [ ] ALL validation commands run and passed
- [ ] NO blocking issues remain
- [ ] User confirmation received
Phase Summary Format
✅ Phase [X] Complete: [Phase Name]
Tasks: [X/X] completed
Reviews: [N] passed
Validations: ✓ All passed
Key outputs:
- [Output 1]
- [Output 2]
Should I proceed to Phase [X+1]: [Next Phase Name]?
Blocked State Handling
When execution is blocked:
⚠️ Implementation Blocked
Phase: [X]
Task: [Description]
Reason: [Specific blocker]
Options:
1. Retry with modifications
2. Skip task and continue
3. Abort implementation
4. Get manual assistance
Awaiting your decision...
Review Handling Protocol
After implementation, handle review feedback:
- APPROVED/LGTM/✅ → proceed to next task
- Specification violation → must fix before proceeding
- Revision needed → implement changes (max 3 cycles)
- After 3 cycles → escalate to user
Context Accumulation
- Phase 1 context = PRD/SDD excerpts
- Phase 2 context = Phase 1 outputs + relevant specs
- Phase N context = Accumulated outputs + relevant specs
- Pass only relevant context to avoid overload
Progress Display
📊 Overall Progress:
Phase 1: ✅ Complete (5/5 tasks)
Phase 2: 🔄 In Progress (3/7 tasks)
Phase 3: ⏳ Pending
Phase 4: ⏳ Pending
Completion Protocol
When all phases complete:
🎉 Implementation Complete!
Summary:
- Total phases: X
- Total tasks: Y
- Reviews conducted: Z
- All validations: ✓ Passed
Suggested next steps:
1. Run full test suite
2. Deploy to staging
3. Create PR for review
Output Format
After phase operations, report:
📍 Phase Execution Status
Phase: [X] - [Name]
Status: [In Progress / Complete / Blocked]
Tasks:
- ✅ [Completed task 1]
- ✅ [Completed task 2]
- 🔄 [Current task]
- ⏳ [Pending task]
Next: [What happens next]
Quick Reference
Phase Boundaries Are Stops
Always wait for user confirmation between phases.
Respect Parallel Hints
Launch concurrent agents when tasks are marked [parallel: true].
Track in TodoWrite
Real-time task tracking during execution.
Update PLAN.md at Phase Completion
All checkboxes in a phase get updated together.
04 同仓库更多 Skills
xiaolai-methodology
以李笑来(Li Xiaolai)的思考方式、方法论与表达风格,为用户做全方位 1-on-1 教练指导,覆盖创业、赚钱、投资、学习、自学、读书写作、用 AI、以及人生心态与自我管理。当用户提到 "李笑来"、"笑来"、"笑来老师"、"扮演李笑来"、"找笑来聊聊"、"笑来怎么看"、"xiaolai"、"把时间当作朋友"、"通往财富自由之路"、"定投改变命运"、"韭菜的自我修养"、"自学是门手艺"、"让时间陪你慢慢变富"、"七年就是一辈子"、"元认知"、"活在未来"、"复利/长期主义"、"践行"、"痴迷改进"、"个人商业模式/出售时间"、"睡后收入"、"定投"、"财富自由"、"操作系统升级"、"最少必要知识"、"北极星提示词"、"用 AI 该怎么学/思考" 等任一触发词,或当用户带着关于创业/赚钱/投资/学习/职业/重大人生决策的困惑、希望被人用笑来式的「逻辑 + 大白话 + 类比 + 重新定义 + 践行导向」来点醒和指导时,自动进入李笑来教练模式。
zread
Produce and consume a wiki-style knowledge base for a code repository via the `zread` CLI and its on-disk output under `./.zread/wiki/`. Use this skill whenever the user wants to understand, onboard onto, explore, summarize, map, or get an overview of an unfamiliar codebase; asks for architecture docs, a project wiki, a repo walkthrough, module/package explanations, or "what does this repo do"; wants to generate, regenerate, resume, browse, or serve code documentation locally; or mentions zread / zread.ai directly. Also use it proactively before diving into a large unknown repo — if `./.zread/wiki/current` exists, read the generated pages instead of crawling source file-by-file; if it doesn't, consider offering to run `zread generate`. The trigger is the intent (understand a codebase through generated docs), not the literal word "zread".
youtube
Launch the YouTube Creator Agent to auto-generate videos. Use when: user types /youtube, wants to create YouTube videos, generate video content, or upload to YouTube. Keywords: youtube, video, 视频, 创作视频, 发布视频
youtube-script-sync
Sync, update, or optimize Tony's GetNote knowledge base `Youtube视频逐字稿` by turning each parent recording or first-person idea note into an idempotent child note containing a directly recordable YouTube long-video or Shorts script. Use when the user says “更新/同步/优化 Youtube视频逐字稿知识库”, or says “更新/同步/优化该知识库” while this knowledge base is the active context.
writing-skills
How to write SKILL.md files that trigger reliably and teach effectively. Use when creating, improving, or reviewing Claude Code skills.
writing-rules
How to write .claude/rules/ files that Claude actually follows. Use when creating, improving, or reviewing project rules.