AgentSkillsCN

implement

强制触发条件：‘要继续：使用 Skill 工具调用“实现”’、‘调用“实现”’、‘带参数实现’、添加问题 ID、‘执行计划’、‘构建这个’、‘按照此计划编码’、计划中需包含“任务”章节。从“带参数 X”中提取问题 ID 作为技能参数。切勿手动实现——务必使用此技能。

SKILL.md
--- frontmatter
name: implement
description: "MANDATORY trigger on: 'To continue: use Skill tool to invoke implement', 'invoke implement', 'implement with arg', beads issue ID, 'execute the plan', 'build this', 'code this plan', plan with Tasks section. Extract issue-id from 'with arg X' as skill argument. NEVER manually implement - ALWAYS use this skill."
argument-hint: "[issue-id] [--fresh]"

Implement

Execute plan from explore, tracked via beads.

Flags

  • •--fresh: Clear context first (EnterPlanMode → summarize progress → user approves → continue)

Steps

  1. •Find plan:
    • •arg → bd show <arg>
    • •no arg → bd list --status in_progress (resume existing)
    • •still nothing → bd ready (pick next unblocked)
    • •No issue? → "Nothing to do. Run /explore first"
  2. •If --fresh: EnterPlanMode, summarize completed/remaining from notes, get approval
  3. •Mark in_progress: bd update <id> --status in_progress
  4. •CHECK FOR OPEN QUESTIONS in plan:
    • •If notes has unresolved questions → resolve via AskUserQuestion FIRST
  5. •FOR EACH TASK (MANDATORY subagent dispatch):
    code
    a. Task tool → implementer subagent (paste full task text)
   - Implementer MUST use TDD: failing test → minimal code → pass
b. Task tool → spec reviewer subagent → if issues: fix → re-review
c. Task tool → quality reviewer subagent → if issues: fix → re-review
d. Update beads notes with progress
    • •NEVER implement tasks yourself - ALWAYS dispatch via Task tool
    • •If task fails → use Skill tool to invoke debugging
    • •TDD is mandatory unless explicitly building throwaway prototype
  6. •EARLY VERIFICATION: After FIRST task, run build/check to catch design flaws
  7. •Update notes after each task: bd update <id> --notes "COMPLETED: X. NEXT: Y"
  8. •Context getting stale? → suggest /implement --fresh to user
  9. •Final → bd close <id> + use Skill tool to invoke finishing-branch

Subagent Prompts

  • •subagent-driven-development/implementer-prompt.md
  • •subagent-driven-development/spec-reviewer-prompt.md
  • •subagent-driven-development/code-quality-reviewer-prompt.md

Key: paste full task text + context into prompt. Don't make subagent read files.