Token导航 LogoToken导航TokenDH.com
研究检索只读github未标认证来源可访问许可证需确认审计通过

plan-first首先计划

Agent Skill

plan-first 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

288

周安装

12

GitHub Stars

公开资料未说明

下载量

96
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:plan-first(首先计划)
来源仓库:https://github.com/xixiaofinland/agent-skills
仓库路径:skills/plan-first
安装命令:
npx skills add https://github.com/xixiaofinland/agent-skills --skill plan-first
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/xixiaofinland/agent-skills --skill plan-first

简介

plan-first 用于查找、检索和筛选相关信息。

  • 适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。
  • 通过 npx skills add 命令从指定仓库安装并使用。
  • 安装前需确认权限范围和维护状态,注意可能触发联网或命令执行。
  • 建议结合原始 README 核验具体用法和功能细节。

SKILL.md

Plan-First Workflow

Rules

  • NEVER write code, create files, or run commands before a TODO.md is approved.
  • NEVER assume missing information. Ask instead.
  • NEVER skip steps. Follow phases in order.
  • NEVER go off-plan. If new work is discovered, add it to TODO.md and ask for approval before doing it.

Phase 1 — Analyze the Project

Read the project silently before asking anything. Check:

  1. Directory structure (top 2 levels)
  2. package.json, pubspec.yaml, go.mod, requirements.txt, Cargo.toml, pom.xml, or equivalent
  3. Existing dependencies and their versions
  4. Build system and scripts (Makefile, scripts/, CI config)
  5. README.md or README.*
  6. Any existing TODO.md, TASKS.md, .todo, or open issue files

Do not output analysis results unless directly relevant to your questions.


Phase 2 — Ask Clarifying Questions (One Round Only)

After analysis, identify gaps that would block correct implementation.

  • Ask at most 5 questions in a single message.
  • Only ask what is critical and cannot be inferred from the codebase.
  • Number the questions.
  • Do not ask about things already answerable from the project files.
  • Do not split into multiple rounds — this is your only chance to ask.

Example format:

Before I create the plan, I need a few things clarified:

1. Should the new endpoint require authentication?
2. Is there a preferred database (the project has both SQLite and Postgres configs)?
3. Should existing tests be updated, or only new ones added?

Wait for the user's response before proceeding.


Phase 3 — Create TODO.md

Using the analysis and the user's answers, write a TODO.md file in the project root.

TODO.md Structure

# TODO

## Goal

One sentence describing what will be built or fixed.

## Tasks

### 1. <Phase Name>

- [ ] <Concrete, measurable action>
- [ ] <Concrete, measurable action>

### 2. <Phase Name>

- [ ] <Concrete, measurable action>
- [ ] <Concrete, measurable action>

## Notes

Any constraints, decisions, or known risks recorded here.

Requirements

  • Tasks must be small and independently verifiable (one logical change each).
  • Order tasks by dependency (prerequisites first).
  • Each task must be checkable as done/not done.
  • No vague items like "fix things" or "improve code".

After writing the file, show the full contents to the user and ask:

I've created TODO.md. Does this plan look correct?
Reply YES to start, or tell me what to change.

Phase 4 — Revision Loop (if needed)

If the user requests changes:

  1. Ask targeted follow-up questions to resolve the disagreement.
  2. Rewrite TODO.md.
  3. Show the updated plan and ask for approval again.

Repeat until the user approves.


Phase 5 — Execute the Plan

Once approved:

  1. Work through tasks in order, one at a time.
  2. After completing each task, mark it done in TODO.md:

- Change - [] to - [x]

  1. State which task you are starting before you begin it.
  2. Do not start the next task until the current one is complete.
  3. Do not perform any work not listed in TODO.md.

If you discover that an unlisted task is required:

  • Stop.
  • Add it to TODO.md under a ## Discovered Tasks section.
  • Tell the user what was found and why it is needed.
  • Ask for approval before continuing.

When all tasks are marked [x], write:

All tasks in TODO.md are complete.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

32.82%
按下载量换算32

Claude

31.37%
按下载量换算30

Cursor

17.85%
按下载量换算17

Gemini CLI

9.98%
按下载量换算10

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills