the-agency-ai/.claude/skills/crawl-sites

<!-- Flag #62/#63: allowed-tools removed. Inherits Bash(*) from .claude/settings.json. Restricting to specific subcommand patterns at the skill level silently blocks agents on permission prompts the agent cannot see — see dispatch #171 for the devex incident that surfaced this trap. -->

Crawl Sites

Crawl configured sites and extract structured content using the configured crawler engine.

Arguments

• $ARGUMENTS: Optional flags:
  • --site <name> — crawl a specific configured site (default: all)
  • --output <path> — output directory for extracted content
  • --dry-run — show what would be crawled without executing
  • --diff — show changes since last crawl

How to Execute

Step 1: Resolve Provider

Read the crawl provider from claude/config/agency.yaml under crawl.provider.

# agency.yaml
crawl:
  provider: "playwright"  # or "wget", "scrapy", "webfetch"
  sites:
    - name: "docs"
      url: "https://docs.example.com"
      patterns: ["/**/*.html"]
    - name: "blog"
      url: "https://blog.example.com"
      patterns: ["/posts/*"]

The provider maps to a tool: ./claude/tools/crawl-{provider}

Step 2: Check Provider Tool Exists

Verify ./claude/tools/crawl-{provider} exists and is executable. If not:

• For webfetch provider: use the built-in WebFetch tool directly (no external tool needed)
• List available crawl tools: ls ./claude/tools/crawl-*
• Tell the user which providers are available

Step 3: Read Site Configuration

Read the crawl.sites array from agency.yaml. Each site entry has:

• name — identifier for the site
• url — base URL to crawl
• patterns — URL patterns to include

If --site specified, filter to that site only.

Step 4: Dispatch to Provider

For each site, execute: ./claude/tools/crawl-{provider} {url} {patterns} {output}

Or for webfetch provider, use the WebFetch tool directly with each URL.

Step 5: Report

Show the user:

• Pages crawled per site
• Content extracted (file count, total size)
• Any errors or skipped pages
• If --diff, show what changed since last crawl

Provider Contract

Each crawl-{provider} tool must accept:

• Positional: base URL
• --patterns — comma-separated URL patterns
• --output — output directory
• --dry-run — list URLs without fetching

Error Handling

• If no provider configured, default to webfetch (uses built-in WebFetch)
• If site config missing, suggest adding the crawl.sites section to agency.yaml
• Rate limit appropriately — respect robots.txt
• Report unreachable URLs without failing the entire crawl