AgentSkillsCN

implementation-planning

在当前计划成果中,完整产出执行计划,涵盖文件差异、工作流、阶段性任务,以及各道关口。适用于 /plan 实施阶段时使用。

SKILL.md
--- frontmatter
name: implementation-planning
description: Produce the execution plan in the current plan artifact including file deltas, workstreams, phases, and gates. Use during /plan Implementation stage.

/implementation-planning

Purpose

Create the Implementation Plan section for build execution. Run as a sub-agent and return a draft section to the orchestrator; do not write the plan artifact directly.

Required outputs

  • Exhaustive file deltas with change type, explicit owner (existing agent/WS), and rationale.
  • Workstreams with dependencies, merge points, and explicitly owned files.
  • An integration / merge points checklist (what gets integrated, how, and what gates run).
  • Enforce the workstream file-ownership rule: each file delta is owned by exactly one workstream until an explicit merge point.
  • Phases and tasks mapped to workstreams/owners.
  • Evidence-based exit criteria per phase.
  • Build-time gates for each phase.
  • Test Plan including at least a minimal test matrix (risk -> test type -> where it runs).
  • Rollout/Deployment steps (even minimal) and an explicit rollback trigger + rollback steps.
  • Draft section content for ## Implementation Plan

Anti-placeholder rule (hard rule)

  • Do not use placeholder language like “run smoke tests” or “add gates” without naming:
    • the gate name(s),
    • where they run (CI vs local vs deployed), and
    • the entrypoint/command (or test runner/target) and what “green” means.
  • If specifics are truly unknown, convert them into either:
    • an explicit Decision boundary (A/B/C) with a recommended default, OR
    • a DR-backed Defer with an explicit trigger.

Q/A wrapper

  • The orchestrator must run the inline Q/A loop to confirm user understanding and agreement on scope, ownership, and gates.

Ownership policy

  • Prefer existing agents from the plan's ## Context Snapshot or the user-provided roster.
  • If no roster is available, return Questions asking for the available agents/owners before assigning work.
  • Do not default all ownership to the user unless the user explicitly requests it.

Parallel workstreams / multi-agent contract (hard rule)

  • If the user requests multiple agents or parallel workstreams:
    • Ensure the draft includes an explicit agent roster (names/handles) and assigns ownership for each workstream and file-delta cluster.
    • If the roster is missing, ask a single focused question to obtain it (do not guess).
    • The orchestrator should mirror this roster into ## Context Snapshot (call this out explicitly in Notes if needed).

Sub-agent output contract

Return a single block in this shape:

md
DraftSection:
<exact section content for ## Implementation Plan (must include the section header)>

Checklist:
- <criterion>: Pass | Fail

Questions:
- <if blocked>

Notes:
- <optional risks/assumptions/tests updates>

Malformed output handling

  • If you cannot produce the exact section header or required fields, return Questions explaining what is missing and leave DraftSection as N/A.

User experience rule (no "go read the plan")

  • When asking the user to confirm plan readiness, paste the key parts directly in the chat response:
    • File deltas (owned)
    • Workstreams + owned files + merge points
    • Phase list with exit criteria
    • Test matrix + rollback

Output template (required for PlanTier: Full)

Use this exact structure so the /plan validator can deterministically check executability:

md
## Implementation Plan
### Agent roster (required for PlanTier: Full)
- <agent/owner>: <responsibilities>

### File Deltas (exhaustive) + rationale
- path/to/file.ext - change type (create/modify/delete) - owner (WSx / agent) - rationale

### Workstreams + merge points
- WS1: <name>
  - Owner:
  - Depends on:
  - Review gates (named):
    - G-...
  - Owns files:
    - path/to/file.ext
  - Merge point / integration step:

### Phases + tasks + exit criteria
#### Phase 1: <name>
- Owner(s):
- Depends on:
- Tasks (by owner):
  - Owner: <agent>
    - [ ] Task
- Exit criteria (evidence):
- Gates (named):
  - G-...

### Review gates (named + definitions)
- G-...:
  - Where it runs: CI | Local | Deployed
  - Entry point / command:
  - Green means:

### Merge points -> required gates
- MP1: <merge point>
  - Blocks on:
    - G-...

### Test Matrix
- Area/component - risk - test type - where it runs

### Test plan (CI vs deployed)
- CI:
  - ...
- Deployed environment:
  - ...

### Rollout / Rollback
- Rollout:
- Rollback trigger:
- Rollback steps: