clawfetch（clawfetch 命令行）

clawfetch (OpenClaw Skill)

Use the published clawfetch npm CLI to convert a single web page into normalized markdown with a small metadata header.

This skill is a thin wrapper around the public clawfetch package:

• It does not vendor the clawfetch source code.
• It does not perform any hidden git clones.
• It only installs the clawfetch npm package in this skill directory via

a separate, explicit bootstrap step.

Typical use cases for agents:

• Fetch a WeChat / blog / docs article into markdown for later indexing.
• Fetch a GitHub repository README (raw) and store it into a knowledge base.
• Fetch a Reddit thread as a structured "post + comments" markdown document.

The CLI emits a header like:

--- METADATA ---
Title: ...
Author: ...
Site: ...
FinalURL: ...
Extraction: readability|fallback-container|body-innerText|github-raw-fast-path|reddit-rss
FallbackSelector: ...   # only when not readability
--- MARKDOWN ---
<markdown>

Installation (skill-level)

This skill assumes:

• node / npm are available in the OpenClaw container.
• Network access to npm to install the clawfetch package and its dependencies.

1. Install the skill into your workspace

Use the OpenClaw CLI to pull the skill from ClawHub into your active workspace:

openclaw skills install clawfetch

This creates a directory similar to:

~/.openclaw/workspace/skills/clawfetch

At this point only the skill wrapper files exist (SKILL.md, README, manifest, bootstrap_deps.sh, etc.). The actual npm package with the scraper logic is not installed yet.

2. Bootstrap the npm CLI once (required)

From the skill directory, run the bootstrap script to install the npm package locally:

cd ~/.openclaw/workspace/skills/clawfetch
bash bootstrap_deps.sh

This script is small and reviewable; it only calls:

npm install clawfetch@0.1.7

Key points:

• OpenClaw's skills install does not currently auto-run this bootstrap

step. Agents or humans must call bootstrap_deps.sh explicitly.

• The install happens inside the skill directory. No global npm state is

modified.

After this completes, the CLI entrypoint is available at:

~/.openclaw/workspace/skills/clawfetch/node_modules/clawfetch/clawfetch.js

Runtime usage (for agents)

After the bootstrap step, agents can invoke the CLI from this directory as:

node node_modules/clawfetch/clawfetch.js <url> [--max-comments N] [--no-reddit-rss]

Recommended patterns:

• General articles / docs

- Use the default mode, which launches headless Chromium via Playwright, then uses Readability + Turndown to extract the main article body.

• GitHub repositories

- Pass the repository URL (e.g. https://github.com/owner/repo). - clawfetch will attempt a raw README fast-path from raw.githubusercontent.com before falling back to browser mode.

• Reddit threads

- Pass the thread URL (e.g. https://www.reddit.com/r/.../comments/.../). - By default, clawfetch will: - Convert the URL to .rss and fetch the Atom feed using a normal desktop browser User-Agent. - Render the first entry as the main post (## Post: ...). - Render subsequent entries as comments (### Comment by ...), limited by --max-comments (default 50; 0 means no limit). - Use --no-reddit-rss to force browser scraping instead of RSS.

Safety / non-suspicious behaviour

• This skill does not:

- Clone any git repositories at runtime. - Download arbitrary source trees into the skill directory. - Run hidden package managers beyond the explicit npm install in the bootstrap step.

• All heavy work (Playwright, Readability, Turndown) comes from the

published clawfetch package and its declared npm dependencies.

Agents should treat this skill as a read-mostly web → markdown tool and avoid using it for arbitrary scripting or filesystem operations beyond its intended CLI interface.