设计与多媒体

Code Wiki

试用

Trigger a one-shot workflow that clones (or pulls) a Git repository, generates a structured wiki from it using the `codewiki` CLI, and optionally renders the...

它能做什么

Trigger a one-shot workflow that clones (or pulls) a Git repository, generates a structured wiki from it using the `codewiki` CLI, and optionally renders the result into static MkDocs / VitePress sites with an HTTP preview. Use this skill whenever the user asks to "generate a wiki", "document this repo", "create a code wiki", or invokes `/codewiki <input>`.

技能文档

Skill: Code Wiki

When to trigger

Invoke this skill when the user:

  • Types /codewiki , or
  • Asks in natural language to generate a wiki / documentation from a repository (e.g. "generate a wiki for anthropics/claude-code").

Trigger command

/codewiki 

Dependencies

This skill depends on CodeWiki (FSoft-AI4Code/CodeWiki). The codewiki CLI — provided by that project — must be installed and available on $PATH before the underlying script can run. The script's startup check_dependencies step will fail with exit code 2 and a clear message if codewiki is missing.

If the user has not installed CodeWiki yet, direct them to and have them follow the install instructions there before confirming execution.

When the user requests rendering (--render), two additional host tools are required: python3 (MkDocs) and node + npm (VitePress). The per-renderer dependencies are installed automatically into isolated sandboxes (Python venv / private node_modules) at render time, so the user never needs to run pip install or npm install manually. If either host tool is missing, render_docs.sh exits with a friendly install hint.

Agent execution logic

Follow these steps in order. Do not execute the underlying script until the user has explicitly confirmed.

1. Parse ``

Extract five fields from the user's input. Accept natural-language phrasing and shorthand.

FieldRequiredAccepted formsDefault
repositoryyesgroup/repo, https://github.com/..., git@github.com:...
output dirnoany local path.
instructionsnofree-form text describing focus / constraints(none)
render stacknomkdocs, vitepress, both, or "skip"skip
servenoboolean, optionally with a port numberoff

Detection heuristics:

  • repository: the first token matching ^\w[\w.-]*/[\w.-]+$, or starting with https:// or git@.
  • output dir: a token following words like into, to, under, at, or a path-looking token (./, /, or containing /).
  • instructions: everything remaining after repository and output are removed. Treat quoted strings as a single instruction block.
  • render stack: triggered by keywords like render, mkdocs, vitepress, build site, static site. Map mkdocs / vitepress / both directly. The word "render" alone (without a named stack) should be clarified with the user before defaulting — do not silently pick a stack.
  • serve: triggered by serve, preview, open in browser, live preview. If a port number appears nearby (e.g. "on 3000"), use it; otherwise omit the port and let render_docs.sh default to 8000. Serving implies --render is also set — if --serve is requested but no stack is named, ask the user which stack to render.

2. Present a structured confirmation overview

Render the parsed options back to the user in a compact block before doing anything. Example:

╭─ codewiki — parameters ──────────────────────────────╮
│ Repository   : anthropics/claude-code                │
│   (expanded) : https://github.com/anthropics/        │
│                claude-code.git                       │
│ Output dir   : .                                     │
│ Instructions : "Focus on the auth module; skip       │
│                vendored code."                       │
│ Render stack : both                                  │
│ Serve        : on, port 3000                         │
╰──────────────────────────────────────────────────────╯

If a field was inferred from natural language (not stated literally), mark it with (inferred) so the user can correct it. If Serve is on, always show the URLs that will be served (http://localhost:PORT for mkdocs, http://localhost:PORT+1 for vitepress when --render both) so the user knows what to expect.

3. Prompt for confirmation or modification (STRICT)

Present exactly two actionable paths. Do not run the script yet.

Confirm to execute, or tell me what to change.

  • Reply yes / confirm / go → I will run:
    ./scripts/codewiki.sh -r anthropics/claude-code \
                -o . \
                -i "Focus on the auth module; skip vendored code." \
                --render both \
                --serve 3000
    
  • Reply modify: = → I will update the parameter table and re-prompt.
  • Reply cancel → abort without side effects.

4. Execute (only after confirmation)

Run the underlying script with the exact flags agreed above. Stream its output to the conversation. On completion, report:

  • The path to the generated wiki (typically //wiki).
  • The paths to any rendered static sites (//mkdocs/site, //vitepress/site).
  • The preview URLs if --serve was used.
  • Any warnings from codewiki or render_docs.sh worth the user's attention.

If --serve was used, the script blocks in the foreground serving HTTP. Tell the user explicitly to press Ctrl-C when done previewing — the cleanup trap will tear down the server(s) automatically.

Hard rules

  • Never run codewiki.sh before the user confirms. This skill clones remote repositories — that is a side effect with network and disk cost. Rendering and serving add further disk and long-running process side effects.
  • Never invent a repository when none was provided. If parsing yields no repository, ask the user for one.
  • Never silently pick a render stack. If the user says "render" but does not name mkdocs / vitepress / both, ask which stack they want.
  • Never start --serve without --render. The script will reject this with exit code 1, but call it out during parsing so the user is not surprised.
  • Always show the expanded clone URL when the user supplied a shorthand group/repo form, so they can verify the target before cloning.
  • Always show which ports will be opened when --serve is used.
  • Preserve instruction text verbatim. Do not reword, summarize, or trim it before passing it to -i.

相关技能

Delegate coding, repository analysis, file edits, test runs, or code review to the local Codex CLI without embedding an OpenAI API key. This skill was created to work with ChatGPT/Codex enterprise accounts that may not have the same level of API access. Invokes codex exec with existing ChatGPT/Codex CLI authentication and returns Codex's final output.

6 次安装1 星标

Codegen (codegen.com). Use this skill for ANY Codegen request — searching and reading data. Whenever a task involves Codegen, use this skill instead of calli...

1 次安装

Produce a durable onboarding artifact for a codebase — writes `ONBOARDING.md` (or `docs/ONBOARDING.md` if `docs/` exists) covering a "read this first" minimum, system overview, dependency map (top-level deps + how each is actually used), startup flow (entry points → bootstrap → config), auth flow (or explicit "none detected"), and 5–15 important files — every claim backed by a `file:line` citation. Walks the repo via parallel Explore sub-agents so big projects don't blow context, calls out what makes THIS codebase non-obvious (not generic framework descriptions), refreshes an existing onboarding doc instead of rewriting from scratch, and renders a condensed summary inline. Built for revisiting a project after months away and for new teammates landing in an unfamiliar repo. Use this skill whenever the user says "explain this codebase", "explain the codebase", "onboard me", "give me a tour", "tour this repo", "what does this repo do", "where do I start", "I haven't looked at this in mont

1 次安装

Convert a book, paper, document, documentation site, or code repository into a structured, on-demand agent skill. Use when the user wants to turn a PDF, EPUB, DOCX, a URL, a docs site, or a GitHub repo into a skill they can load later — "make a skill from this book", "turn this paper into a skill", "turn these docs into a skill", "I want an agent that knows this library".

3 次安装

Builds and maintains a global, project-independent knowledge wiki from user-uploaded documents using Karpathy's native LLM Wiki paradigm (zero vectors, zero chunks, single-page units, two-level topic/page index). Invoke when user uploads documents and asks to 'create wiki / build wiki / compile know

Use this skill when you need to publish, fetch, search, list, share, or watch AgentFiles artifacts from Codex, Claude Code, OpenClaw, or other agent runtimes...

17 次安装