跳到正文
cccskills
免费在 GitHub 公开

plan

Create a concise, evidence-backed implementation plan and save it to .agents/plans/Plan.md. Use when the user asks to plan a feature or task.

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md3.2 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

The plan file is .agents/plans/Plan.md: a record of decisions and executable tasks.

1. Understand the request

Restate the goal in your own words; ask when the intent is ambiguous or a trade-off matters. When the direction itself is unsettled, run a grilling session first to settle it. Read the repository instructions (AGENTS.md, README, docs) and any existing plans in .agents/plans/.

Done when the goal is restated and no ambiguity that could change the plan is left unasked.

2. Research

  • Inspect the code, tests, schemas, and configuration the task would change; trace the current behavior through real entry points.
  • Use web search for external knowledge — library or API behavior, version compatibility, security guidance, community best practice. Fetch the authoritative source.
  • Use subagents for wide or parallel exploration — a large or unfamiliar codebase, many candidate locations, independent concerns. Dispatch one read-only subagent per area; demand evidence with file:line references.
  • Follow every claim back to the source that owns it — primary sources (code, official docs, specs), not secondary write-ups. Record facts separately from proposals.

Done when the current behavior is traced and every claim the plan will make carries a citation (file:line for code, URL for web).

3. Draft the plan

Use only sections that carry information:

# Plan: [Feature]

## Goal

What outcome changes, for whom, and why.

## Current State

Relevant behavior, entry points, constraints, and evidence, with
citations (file:line or URL).

## Decisions

Each decision: the choice, the alternatives considered, and why they were
rejected. Decisions describe contracts and intent, not file paths — they
must stay true as the code moves.

## Scope

In scope. Intentionally out of scope.

## Tasks

Ordered, bite-sized tasks with checkboxes. Each task:

- Names the files it touches (`**Files:**`).
- Names the seam at which its tests hook in (`**Seam:**`) — prefer an
  existing seam, and the highest one possible.
- Gives a way to verify it (`**Verify:**` command or check).

- [ ] Task 1: ... (**Seam:** ...; **Files:** ...; **Verify:** ...)
- [ ] Task 2: ...

## Verification

Focused checks and observable acceptance criteria.

## Open Questions

Only unresolved decisions that can change the plan.

Rules:

  • Name concrete files and symbols only where evidence supports them.
  • No placeholders: "TBD", "implement later", or "add error handling" without specifics are plan failures. Each task must be executable from the plan alone.

Done when the self-review passes — every goal maps to at least one task, names, signatures, and paths are consistent across tasks, and no placeholder remains. Fix issues inline.

4. Save

Write the final plan to .agents/plans/Plan.md with the write tool. If the file exists and the new plan supersedes the old one, overwrite it; otherwise, ask the user for a filename.

Done when the plan is saved. End your turn with a short summary and the file path.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

Review the changes since a fixed point (commit, branch, tag, or merge-base) along two axes — Standards (does the code follow this repo's documented coding standards?) and Spec (does the code match what the originating issue/spec asked for?). Runs both reviews in parallel sub-agents and reports them side by side. Use when the user wants to review a branch, a PR, work-in-progress changes, or asks to "review since X".

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

Execute git commits using the Conventional Commits specification. Use this skill whenever the user says "commit", "git commit", "/commit", "stage and commit", "make a commit", "save my changes", or asks to commit specific files or all changes. Also triggers on phrases like "push my changes" (commit first), "commit everything", "commit with message X", or whenever git diff/status suggests uncommitted work the user wants to save. Handles auto-staging, diff analysis, message generation, breaking change detection, batch splitting of unrelated changes, safe pushes, and pre-commit hook failures gracefully. Always use this skill rather than ad-hoc git commit commands.

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

graphify

免费

Use for any question about a codebase, its architecture, file relationships, or project content — especially when graphify-out/ exists, where the question should be treated as a graphify query first. Turns any input (code, docs, papers, images, videos) into a persistent knowledge graph with god nodes, community detection, and query/path/explain tools.

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

grilling

免费

Grill the user relentlessly about a plan, decision, or idea. Use when the user wants to stress-test their thinking, or uses any 'grill' trigger phrases.

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

implement

免费

Implement a piece of work based on a spec or set of tickets.

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

Maintain CHANGELOG.md — add new version entries, keep the Unreleased section, cut a release, or restructure the whole file — following the Keep a Changelog format and Semantic Versioning.

中文简介正在准备中。当前显示原文说明。

nmdra/Dotfiles282026年8月11日 更新

查看 nmdra 的全部技能

报告此技能的问题