CommunityArt & Designgithub.com

BoyLei/CodexSkill_Doc

codex 关于 插件/skill/agent 的说明文档

Works withClaude CodeCodex CLI~Cursor
npx skills add BoyLei/CodexSkill_Doc

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Document Local Extensions

Use this skill to turn a locally installed Codex plugin, app bundle, skill, or related CLI package into a readable operation guide.

Workflow

  1. Identify the target from the user's wording.

    • Accept names such as Token Tracker, Figma, CodeGraph, a skill name, a plugin folder name, a command name, or a local path.
    • If the target is ambiguous, search likely local extension roots first before asking.
  2. Locate the local files.

    • Search ~/.codex/skills for skills.
    • Search ~/.codex/plugins and ~/.codex/plugins/cache for plugins and bundled app assets.
    • Search package or command locations only when the extension exposes a CLI, for example with Get-Command <name> or python -m pip show -f <package>.
    • Prefer exact names first, then normalized variants: spaces, hyphens, underscores, and case-insensitive matches.
  3. Read the highest-signal files first.

    • Skills: SKILL.md, agents/openai.yaml, directly referenced references/*, and relevant scripts.
    • Plugins: .codex-plugin/plugin.json, command markdown files, skill folders, app/MCP metadata, README-like docs if present.
    • CLI packages: console entry points, --help output, command dispatch source, package metadata, and docs.
  4. Extract the practical interface.

    • What the extension does.
    • What commands or tools it exposes.
    • Which files or config it reads and writes.
    • Which operations are read-only and which change local state.
    • Required environment variables, PATH entries, encodings, versions, or restart steps.
    • Common failures and short fixes.
  5. Verify commands when safe.

    • Run read-only checks such as --version, --help, Get-Command, pip show, or listing local files.
    • Do not run setup, install, delete, reset, login, send, publish, or destructive commands unless the user explicitly asks.
    • On Windows, use -LiteralPath and UTF-8-safe PowerShell wrappers when command output may contain Chinese or symbols.
  6. Write the operation guide as Markdown.

    • Put the generated .md in the user's requested folder.
    • If no folder is specified, use the current workspace or an obvious documentation folder.
    • Keep it practical: commands, examples, config paths, and troubleshooting over background explanation.
  7. Maintain 插件简介.md when Codex extensions change.

    • On any plugin, skill, or agent addition or removal, rebuild 插件简介.md in the requested folder or current workspace.
    • Enumerate the current installed state from ~/.codex/skills, ~/.codex/plugins, ~/.codex/plugins/cache, ~/.codex/agents, and ~/.agents/plugins when they exist.
    • Include one compact entry per current plugin, skill, or agent: type, name, local path, and one-sentence function summary.
    • Omit removed extensions instead of keeping stale entries.

Windows Command Patterns

Use these patterns when inspecting local extension files:

Get-ChildItem -LiteralPath 'C:\Users\dl\.codex\skills' -Recurse -Force |
  Where-Object { $_.Name -match '(?i)target-name' } |
  Select-Object FullName,PSIsContainer,Length
rg -n -i "target name|target-name|target_name" 'C:\Users\dl\.codex' `
  -g '!auth.json' -g '!*.sqlite' -g '!*.sqlite-shm' -g '!*.sqlite-wal'

Use these patterns when inspecting a CLI:

Get-Command command-name -ErrorAction SilentlyContinue | Select-Object Source,Version
where.exe command-name
command-name --version
command-name --help

If a Python CLI prints Unicode poorly in Windows PowerShell, run it with:

$env:PYTHONIOENCODING='utf-8'; $env:PYTHONUTF8='1'; command-name --help

Markdown Output Shape

Use this structure unless the target calls for something different:

---
name: target-name
description: One sentence describing when to use this local extension.
---

# Target Name 操作指南

## 本地安装位置

- Extension path:
- Command path:
- Version:

## 功能概览

- ...

## 常用命令

```powershell
...

会修改本机状态的操作

  • ...

读取的数据和配置

  • ...

Windows 注意事项

  • ...

常见排查

  • ...

给 Codex 的使用提示

  • ...

## Safety Rules

- Do not read or print secrets from `auth.json`, token files, browser profiles, environment variables, or credential stores.
- Do not infer behavior only from file names; verify from manifests, docs, source, or harmless command output.
- Do not mutate `~/.codex/config.toml`, plugin caches, PATH, or package installs unless the user asks for that action.
- If an extension has both read-only commands and setup/unsetup commands, document them separately.
- If command output is mojibake, re-read files with `-Encoding UTF8` or use a UTF-8 environment wrapper before concluding the content is corrupt.

Related Skills

notoetannoto/revdiff

Review diffs, files, and docs in a TUI with inline notes, then export structured annotations for scripts and AI tools

community

dpearson2699/swiftui-liquid-glass

Implement, review, or improve SwiftUI Liquid Glass effects for iOS 26+. Covers glassEffect modifier, GlassEffectContainer, glass button styles, glass toolbar, glass tab bar, morphing transitions, translucent material, vibrancy, tinting, interactive glass, ToolbarSpacer, scrollEdgeEffectStyle, backgroundExtensionEffect, and availability gating. Use when asked about Liquid Glass, glass buttons, glassEffect, GlassEffectContainer, GlassEffectTransition, glassEffectID, glassEffectUnion, scroll edge effects, or adopting iOS 26 design.

community

gquthier/demonte-ton-offre

Skill Claude Code : démonte une offre existante et la reconstruit en offre irrésistible (méthode Hormozi $100M Offers + frameworks marché/cible)

community

LpcPaul/tool-scout-skill

AI tool discovery Agent Skill for finding existing software tools across GitHub, npm, MCP servers, Agent Skills, VS Code Marketplace, Open VSX, and web search before building from scratch.

community

SoTuffStuff/vibecode-pro-max-kit-379

Your AI forgets. This remembers. Spec-driven coding harness for vibecoders, product owners, CEOs and real builders — self-improving context memory, 12 agents, 32 skills. Kills context rot, ships features, not spaghetti. Claude Code & Codex. Any stack. 30 seconds

community

FransDevelopment/open-agent-trust-registry

The open root-of-trust for agent identity on the internet. A public, federated registry of trusted attestation issuers — the agent runtimes authorized to vouch for agents acting on behalf of humans. Services verify agent attestations against this registry to determine if the issuing runtime is legitimate.

community