fetch

name	fetch
description	Fetch content from a URL and save it to a file. Use this when you need to retrieve web content (documentation, API responses, articles). Content is saved to ~/workspace/.fetch_cache/ and you can read it with the Read tool. For complex research across multiple sources, use the web-researcher subagent instead.
version	1.0.0

Fetch Skill

Downloads web content and saves it to a local file for you to read.

Usage

~/.claude/skills/fetch/fetch.py <url>

What It Does

1. Fetches the URL with a realistic browser user-agent
2. Converts HTML to clean Markdown (or formats JSON)
3. Saves to ~/workspace/.fetch_cache/<hash>.md with metadata
4. Returns the file path for you to read

Output File Format

---
url: https://example.com/docs
final_url: https://example.com/docs  # after redirects
fetched_at: 2026-03-21T14:32:00+00:00
content_type: text/html
detected_type: html
size_bytes: 45230
status_code: 200
---

# Page Title

Content in markdown format...

Examples

Fetch documentation

~/.claude/skills/fetch/fetch.py https://docs.python.org/3/library/json.html
# Returns: File path, then use Read tool to access

Fetch API schema

~/.claude/skills/fetch/fetch.py https://api.example.com/openapi.json

Reading the Content

After fetching, use the Read tool to access the content:

# Read entire file
Read: ~/workspace/.fetch_cache/a1b2c3d4e5f6.md

# Read first 100 lines only (for large files)
Read: ~/workspace/.fetch_cache/a1b2c3d4e5f6.md (limit: 100)

# Paginate through large files
Read: ~/workspace/.fetch_cache/a1b2c3d4e5f6.md (offset: 100, limit: 100)

When to Use This vs web-researcher

Use /fetch when...	Use web-researcher subagent when...
You need one specific URL	You need to research a topic
Quick lookup	Multiple sources needed
You know exactly what page has the info	You need to search first
Small to medium content	Synthesis across sources

Cache Location

All fetched content is stored in:

~/workspace/.fetch_cache/
├── a1b2c3d4e5f6.md
├── f7e8d9c0b1a2.md
└── ...

Files persist across sessions. Delete manually if needed.