Web Documentation Resolver

Resolve query or URL inputs into compact, high-signal markdown using an intelligent cascade.

Platform Tool Mapping

Cascade Resolution Strategy

URL Inputs

1. llms.txt: Probe https://origin/llms.txt first (free, structured)
2. Fetch: Use platform fetch tool for markdown content
3. Search fallback: Find cached/mirrored versions if fetch fails

Query Inputs

1. Search first: Use platform search tool (fast, free)
2. Fetch top results: Get markdown from promising results

Python Script

python scripts/resolve.py "https://docs.rust-lang.org/book/"  # URL
python scripts/resolve.py "Rust async programming"           # Query
python scripts/resolve.py "query" --json --max-chars 4000    # Options

Direct Tool Usage

# opencode
webfetch https://example.com/llms.txt
webfetch --format markdown https://docs.rust-lang.org/book/
websearch "Rust book documentation"

# claude code (MCP)
WebFetch(url="https://example.com/llms.txt")
WebFetch(url="https://docs.rust-lang.org/book/")
WebSearch(query="Rust book documentation")

# blackbox
web_fetch(url="https://example.com/llms.txt", prompt="Extract all content")
web_search(query="Rust book documentation")

Best Practices

• Check /llms.txt first for structured documentation
• Use specific queries: "tokio spawn vs spawn_blocking difference" > "tokio"
• Add year to queries for current info: "Rust async 2026"
• Prefer official docs; try mirrors if primary fails

Quality Indicators

Good: Code examples, API signatures, config samples, version info, clear structure Poor: Boilerplate, paywalls, login requirements, heavy ads

Error Handling

• Cascade fallback on provider failures
• Log errors for debugging
• Search fallback when direct fetch fails

Testing

cd .agents/skills/web-doc-resolver
python -m pytest tests/ -v

Files

scripts/resolve.py - Main implementation | tests/test_resolve.py - Unit tests | reference.md - Detailed reference