AGENT SKILL

implementation-plan

Turns requirements into a short implementation plan of small, verifiable steps, each with the files it touches and the test scenarios it covers. Use after requirements are clear and before coding, or when asked to plan, break down or estimate a change.

License 许可
GPL-3.0
最近更新
2026-11-12
Stars
★ 0
Forks
0

01 前言 Front Matter

name
implementation-plan
description
Turns requirements into a short implementation plan of small, verifiable steps, each with the files it touches and the test scenarios it covers. Use after requirements are clear and before coding, or when asked to plan, break down or estimate a change.

02 资源文件 1

03 SKILL.md 正文

Implementation plan

Goal: a plan anyone can follow step by step, where each step leaves the project working and tested. Describe how; the what comes from the requirements.

1. Start from the requirements

Read the requirements document (see the requirements skill). If there is none and the change is more than trivial, write it first. Every step of the plan must trace back to a story or acceptance criterion.

2. Explore the code

Find out how things work today before deciding anything:

  • the files and modules the change touches;
  • existing patterns to follow (similar features, naming, tests);
  • the project's conventions (AGENTS.md, CLAUDE.md, CONTRIBUTING.md), its spec (docs/spec.md) and past decisions (docs/adr/).

If the exploration is large and a read-only exploration subagent is available, delegate it and ask for a summary of files, patterns and risks. Do not let it decide anything.

3. Decide the approach, with the user

When there is more than one reasonable way, show the options in a few lines each (what it is, what it costs, what it risks), recommend one and ask. Do not ask when one option is clearly better; state it and move on.

Prefer the simplest approach that meets the requirements, and weigh options with the design-principles skill. Reuse what exists before adding something new. Any new dependency needs the user's approval. A decision that is hard to reverse gets an ADR (see the docs skill).

4. Break it into steps

Size the plan to the change: a small change gets one or two steps. Steps organize the work and its review; they do not decide the commits (see the git-workflow skill). Never add a step whose only purpose is to commit, such as committing the plan alone or marking it done.

Each step:

  • is small enough to review on its own (one idea, a handful of files);
  • leaves the project building and its tests passing;
  • lists the files it touches and the scenarios it covers (these become the tests in the tdd skill);
  • says how to verify it (command to run, behavior to check).

If the change alters behavior, include a step to update the docs it makes false (spec, README, CHANGELOG).

Order: what others depend on first; the main path before edge cases; risky or unknown parts early, so surprises appear when they are cheap.

5. Name the risks

List what could go wrong or is still unknown (migrations, breaking changes, performance, security) and how the plan handles each. Skip what does not apply.

Output

Fill assets/template.md and save it where the project keeps plans; if it has none, use docs/plans/<feature-name>.md. Reply in a few lines with the approach, the number of steps and anything the user must decide.

If the project's rules ask for a plan document only for large changes, give a small change its plan in the reply instead of a file, filled from the same template.

Do not start coding until the user approves the plan.

04 同仓库更多 Skills

gustavoSaboia97
gustavoSaboia97/casulo

{name=null}

Describe what {name} does and when the agent should use it

★ 0
gustavoSaboia97
gustavoSaboia97/casulo

tdd

Develops with TDD driven by BDD scenarios (Given/When/Then), writing only the functional, regression and unit tests that protect real behavior. Use when implementing a feature, fixing a bug, or when asked for tests, TDD, BDD, regression tests or coverage.

★ 0
gustavoSaboia97
gustavoSaboia97/casulo

security-audit

Audits code for exploitable security problems (injection, broken access control, leaked secrets, unsafe input handling, vulnerable dependencies) and reports only verified findings with severity and fix. Use when asked for a security review or audit, before a release, or when a change touches authentication, permissions, secrets, user input, file paths, commands or network calls.

★ 0
gustavoSaboia97
gustavoSaboia97/casulo

requirements

Turns a request or idea into clear, testable requirements, with scope, user stories and Given/When/Then acceptance criteria. Use when starting a new feature or project, when a request is vague, or when asked for requirements, user stories, acceptance criteria or a feature spec.

★ 0
gustavoSaboia97
gustavoSaboia97/casulo

refactoring

Restructures existing code without changing its behavior, in small steps with the tests green after each one. Use when asked to refactor, clean up, simplify, rename, extract, move or reorganize code, or to work on items recorded as "Refactor later".

★ 0
gustavoSaboia97
gustavoSaboia97/casulo

git-workflow

Creates branches, small Conventional Commits and pull request descriptions, safely. Use when committing, writing a commit message, creating a branch, pushing, opening or describing a pull request, or resolving a merge conflict.

★ 0