AgentSkillsCN

plan-feature

起草或优化某项功能的完整实施方案。在过程中适时提出澄清性问题，并将方案记录于`.claude/workflow/plan.md`文件中。

SKILL.md
--- frontmatter
name: plan-feature
description: Draft or refine a complete implementation plan for a feature. Asks clarifying questions, writes `.claude/workflow/plan.md`.
argument-hint: "<feature brief> [optional: answers / constraints / APPROVED]"
context: fork
agent: general-purpose
allowed-tools: Read, Grep, Glob, Bash(rg *), Bash(git *), Write

ultrathink

You are producing a complete plan that will be executed by implementation subagents. Do NOT modify source code. Only write/overwrite .claude/workflow/plan.md.

Inputs:

  • •$ARGUMENTS contains the feature brief and any answers/constraints.
  • •If .claude/workflow/codebase-map.md exists, use it. Otherwise, quickly infer needed structure from the repo.

Required behavior

  1. •If information is missing, produce a prioritized list of clarifying questions.
  2. •Do not mark the plan as complete until open questions are empty.
  3. •Break work into small tasks (checkboxes). Each task must include concrete file paths (or explicit discovery steps) and verification steps.

Plan structure (write exactly this shape)

Write .claude/workflow/plan.md containing:

Plan: <feature name>

Status

  • •Planning: <IN PROGRESS | COMPLETE>
  • •Approved: <NO | YES>
  • •Last updated: <optional>

Feature brief

(From $ARGUMENTS)

Scope

  • •Goals
  • •Non-goals

Assumptions / constraints

(bullets)

Open questions (must be empty to complete planning)

  • • Q1 ...
  • • Q2 ... (If none, write: “None.”)

Proposed approach

  • •High-level design
  • •Alternatives considered (brief)
  • •Impacted components/files (best guess + pointers)

Implementation tasks

  1. • Task 1 — <short name>
    • •Files: <paths or discovery steps>
    • •Steps: <bullets>
    • •Verification: <tests/commands/checks>
  2. • Task 2 ... (Keep tasks small; prefer 2–10 tasks depending on feature size.)

Verification checklist (Definition of Done)

  • • Relevant unit tests added/updated
  • • Integration/e2e tests (if applicable)
  • • Lint/format passes
  • • No TODOs / debug logs left behind
  • • Docs updated (if needed)
  • • Observability/metrics/logging updated (if applicable)

Risk & rollback

  • •Risks + mitigations
  • •Rollback strategy

Approval

  • •To begin implementation, user must say exactly: APPROVE PLAN
  • •Approved: <NO | YES>
  • •Approval notes: <optional>

Output to orchestrator

After writing plan.md, respond with:

  • •Updated files list
  • •Remaining open questions (if any)
  • •If zero open questions: say “Plan ready for validation: run /validate-plan”
  • •If $ARGUMENTS indicates approval (e.g., includes “APPROVED”): set Approved=YES and say “Implementation allowed: run /implement-feature”