AgentSkillsCN

pm-implement

执行已批准的Beads任务,运行自动化实施后评审,完成评审迭代,随后进行手动QA烟雾测试,并返回进行最终用户审核。

SKILL.md
--- frontmatter
name: pm-implement
description: Execute approved Beads tasks, run automated post-implementation reviews, complete review iteration, then run Manual QA smoke tests and return for final user review.

PM Implement (Strict)

Current Phase

  • IMPLEMENTATION (then POST-IMPLEMENTATION REVIEWS, REVIEW ITERATION, MANUAL QA SMOKE TESTS, and AWAITING FINAL REVIEW)

Preconditions (hard gate)

Before implementation starts, verify:

  1. PRD is approved.
  2. Beads task graph is approved.
  3. Epic ID is known and exists.

If any precondition fails:

  • Stop and ask only for missing prerequisite(s).

Implementation Rules

  • Keep paired support agents available:
    • Senior Engineer for proactive code-level guidance and risk checks.
    • Librarian for external docs/compatibility checks when implementation touches APIs/platform specifics.
  • Execute tasks from ready queue first:
    • bd ready --parent <epic-id> --pretty
  • Claim/start work:
    • bd update <task-id> --claim
  • Keep changes scoped to active tasks.
  • Log meaningful status notes:
    • bd comments add <task-id> "<progress/update>"
  • Close completed tasks:
    • bd close <task-id>

Automatic Dual-Agent Post-Implementation Review

After implementation tasks are complete, automatically run both reviewers:

  1. AGENTS Compliance Reviewer

    • Load prompt from references/agents-compliance.md.
    • Check implementation against repo AGENTS.md rules and workflow constraints.
    • Return violations, file references, severity, and required fix.
  2. Jazz

    • Load prompt from references/jazz.md.
    • Persona: grumpy, nitpicky old fart.
    • Behavior: doubt assumptions, challenge weak logic, call out edge cases and missing rigor.
    • Return concrete defects and demanded fixes.

Run both reviewers in parallel when possible.

Preferred orchestration calls:

  • Spawn compliance reviewer agent from references/agents-compliance.md.
  • Spawn Jazz reviewer agent from references/jazz.md.
  • Wait for both to complete before creating iteration tasks.

Review Output Handling (mandatory)

For every actionable review finding:

  • Add comment on mapped task:
    • bd comments add <task-id> "<reviewer>: <finding>"
  • Create iteration task when code changes are required:
    • bd create --type task --parent <epic-id> --title "Review iteration: <short title>" --description "<fix + DoD>" --labels review,iteration
  • Set dependencies so unresolved findings block completion:
    • bd dep <blocking-task-id> --blocks <blocked-task-id>

Review Iteration Loop

  • Implement all review iteration tasks.
  • Re-run targeted validation for changed areas.
  • Close iteration tasks only when their DoD is met.

Manual QA Smoke Tests (mandatory after automated reviews)

  • After automated reviews and review-iteration fixes, run Manual QA smoke execution:
    • load prompt from references/manual-qa-smoke.md
    • execute discovery/PRD smoke tests for happy, unhappy, and regression coverage
    • run browser-based smoke checks when required by the test plan
  • Preferred orchestration call:
    • spawn Manual QA agent and wait for completion before final handoff
  • If smoke tests fail:
    • create/fill beads follow-up tasks with explicit DoD
    • implement fixes
    • rerun Manual QA smoke tests
  • Continue until smoke tests pass or only user-accepted risks remain.

Final Gate

  • Move to Current phase: AWAITING FINAL REVIEW.
  • Present:
    • implementation summary
    • reviewer findings summary
    • review-iteration changes completed
    • Manual QA smoke-test results
  • Ask user to review results.
  • Do not auto-finish beyond this gate.

Output Requirements (every run)

Always include:

  1. Current phase: <...>
  2. Epic ID
  3. Active/ready tasks
  4. Reviewer status (not started/running/completed)
  5. Iteration task status
  6. Manual QA smoke status (not started/running/passed/failed)
  7. What I need from you next

Invocation

  • Trigger strongly on $pm-implement ....
  • Also trigger as automatic handoff from $pm-beads-plan after second approval.