AGENT SKILL

write-plan

Write a project / change plan to docs/plans/<slug>.md from the canonical Plan template; What, Why, Approach, Work breakdown, Risks, Verification, Out of scope. The full plan doc, not a standalone WBS.

License 许可
MIT
最近更新
2026-10-29
Stars
★ 0
Forks
0

01 前言 Front Matter

name
write-plan
description
Write a project / change plan to docs/plans/<slug>.md from the canonical Plan template; What, Why, Approach, Work breakdown, Risks, Verification, Out of scope. The full plan doc, not a standalone WBS.

02 资源文件 1

03 SKILL.md 正文

write-plan

Author a plan that is short, actionable, and explicit about what we are not doing.

Mode: analytical — judgment inside a fixed frame. Keep the output shape; every verdict cites its evidence; unknowns are marked [TBD] or [VERIFY], never filled in.

When to use

  • A non-trivial change needs a written approach before implementation.
  • A workflow invokes polymath-planning:write-plan.
  • The user asks "what's the plan?".

Inputs

  • Title (required).
  • Motivation: PRD path, problem statement, or ADR link.
  • Constraints: timeline, team, dependencies.

Procedure

  1. Read Plan.md.
  2. Slug from the title: a workflow's ${workflow.slug} wins; otherwise lowercase it, replace each run of characters outside a-z0-9 with -, trim - at the ends, and cut to ≤ 60 characters at a -. An existing file at the target path is updated in place (show the diff), never duplicated.
  3. Summary-first checkpoint. Before writing the document, show a one-screen skeleton — the What in one sentence, the 3–5 approach phases, the locked decisions you'll honor, the top risk — and confirm direction. Cheaper to redirect a skeleton than a finished plan. Unattended run (workflow step, subagent): nobody can confirm — skip the wait, write the plan, and open the step summary with the skeleton.
  4. Draft docs/plans/<slug>.md:
    • What — outcome in one paragraph.
    • Why — link the PRD/ADR; quote the constraint that made this necessary.
    • Approach — 3–5 phases, not tasks.
    • Locked decisions — decisions already made (spike/ADR/owner, with source) that the plan must not relitigate.
    • Work breakdown — numbered, each step independently completable.
    • Risks — at least one risk with a mitigation.
    • Verification — observable signals or artifacts.
    • Deferral registry — wanted-but-deferred work, each entry with the condition that reopens it (never a bare "later").
    • Out of scope — explicit non-goals.
  5. Stop. Do not start implementation in the same skill turn.

Quality bar

  • Plan ≤ 60 lines total. If it grows past that, decompose the scope.
  • Every step in the work breakdown is observable from the outside (a file exists, a test passes, a deploy goes out).
  • The "Out of scope" section has at least one entry.
  • Every deferral names its revisit condition; every locked decision names its source.

Output

  • File: docs/plans/<slug>.md.
  • Summary listing the phase count and risk count.

04 同仓库更多 Skills