name: fetcher
description: >
Fetch web pages, PDFs, and documents with automatic fallbacks and content extraction.
Use when user says "fetch this URL", "download this page", "crawl this website",
"extract content from", "get the PDF", or provides URLs needing retrieval.
allowed-tools: Bash, Read
triggers:
- fetch this URL
- download page
- crawl website
- extract content from
- get the PDF
- scrape this site
- retrieve document
metadata:
short-description: Web crawling and document fetching CLI

Fetcher - Web Crawling

Fetch web pages and documents with automatic fallbacks, proxy rotation, and content extraction.

Self-contained skill - auto-installs via uvx from git (no pre-installation needed).

Fully automatic - Playwright browsers are installed on first run for SPA/JS page support.

Simplest Usage

# Via wrapper (recommended - auto-installs)
.agents/skills/fetcher/run.sh get https://example.com

# Or directly if fetcher is installed
fetcher get https://example.com

Common Commands

./run.sh get https://example.com                   # Fetch single URL
./run.sh get-manifest urls.txt                     # Fetch list of URLs
./run.sh get-manifest - < urls.txt                 # Fetch from stdin

Common Patterns

Fetch a single URL

fetcher get https://www.nasa.gov --out run/nasa

Outputs to run/nasa/:
- consumer_summary.json - structured result
- Walkthrough.md - human-readable summary
- downloads/ - raw content files

Fetch multiple URLs

# From file (one URL per line)
fetcher get-manifest urls.txt --out run/batch

# From stdin
echo -e "https://example.com\nhttps://nasa.gov" | fetcher get-manifest -

ETL mode (full control)

fetcher-etl --inventory urls.jsonl --out run/etl_batch
fetcher-etl --manifest urls.txt --out run/demo

Check environment

fetcher doctor                    # Check dependencies and config
fetcher get --dry-run <url>       # Validate without fetching
fetcher-etl --help-full           # All options
fetcher-etl --find metrics        # Search options

Output Structure

run/artifacts/<run-id>/
├── results.jsonl              # Fetch results per URL
├── consumer_summary.json      # Summary stats
├── Walkthrough.md             # Human-readable summary
├── downloads/                 # Raw files (HTML, PDF, etc.)
├── text_blobs/                # Extracted text
├── markdown/                  # LLM-friendly markdown
├── fit_markdown/              # Pruned markdown for LLM input
├── junk_results.jsonl         # Failed/junk URLs
└── junk_table.md              # Quick triage table

Content Extraction

Enable markdown output

export FETCHER_EMIT_MARKDOWN=1
export FETCHER_EMIT_FIT_MARKDOWN=1  # Pruned for LLM input
fetcher get https://example.com

Rolling windows (for chunking)

export FETCHER_DOWNLOAD_MODE=rolling_extract
export FETCHER_ROLLING_WINDOW_SIZE=6000
export FETCHER_ROLLING_WINDOW_STEP=3000
fetcher get https://example.com

Advanced Features

HTTP caching

# Cache enabled by default
fetcher get https://example.com

# Disable cache for fresh fetch
fetcher get https://example.com --no-http-cache

PDF discovery

# Auto-fetch PDF links from HTML pages
export FETCHER_ENABLE_PDF_DISCOVERY=1
export FETCHER_PDF_DISCOVERY_MAX=3
fetcher get https://example.com

Proxy rotation (rate-limited sites)

export SPARTA_STEP06_PROXY_HOST=gw.iproyal.com
export SPARTA_STEP06_PROXY_PORT=12321
export SPARTA_STEP06_PROXY_USER=team
export SPARTA_STEP06_PROXY_PASSWORD=secret
fetcher-etl --inventory urls.jsonl

Brave/Wayback fallbacks

# Enable alternate URL resolution
export BRAVE_API_KEY=sk-your-key
fetcher-etl --use-alternates --inventory urls.jsonl

Python API

import asyncio
from fetcher.workflows.web_fetch import URLFetcher, FetchConfig, write_results
from pathlib import Path

async def main():
    config = FetchConfig(concurrency=4, per_domain=2)
    fetcher = URLFetcher(config)
    entries = [{"url": "https://www.nasa.gov"}]
    results, audit = await fetcher.fetch_many(entries)
    write_results(results, Path("artifacts/nasa.jsonl"))
    print(audit)

asyncio.run(main())

Single URL helper

from fetcher.workflows.fetcher import fetch_url

result = await fetch_url("https://example.com")
print(result.content_verdict)  # "ok", "empty", "paywall", etc.
print(result.text)             # Extracted text

FetchResult Fields

Environment Variables

Troubleshooting

Run fetcher doctor to check environment and dependencies.

SPA/JavaScript Page Support

Fetcher automatically falls back to Playwright for known SPA domains. If a page returns thin/empty content:

1. Check if used_playwright: 1 in consumer_summary.json
2. If not, the domain may need to be added to SPA_FALLBACK_DOMAINS in fetcher source
3. Force fresh fetch with FETCHER_HTTP_CACHE_DISABLE=1