| name | web-discovery |
| description | Plan and run current public-web discovery for docs, releases, issues, and community evidence without guessing or over-browsing. |
| origin | repo-local web research workflow |
Web Discovery
Use this skill when the user needs current public-web research, release/status checks, recommendations, or evidence from multiple sites.
Workflow
- Search the repo, provided links, and known local docs first.
- Prefer official docs, release notes, source repositories, standards pages, and issue trackers before secondary coverage.
- Split broad questions into small queries with exact product names, versions, dates, errors, and domain filters where possible.
- Stop widening scope once the answer is supported by enough primary evidence.
- Hand the gathered sources to
source-audit before making strong claims or recommendations.
Rules
- Separate official statements, maintainer comments, benchmarks, and user anecdotes.
- Use exact dates for unstable facts.
- Do not paste PATs, passwords, internal URLs, or private repo details into external search tools.
- For high-stakes or version-sensitive facts, keep the search set biased toward primary sources.
Output
- A short evidence list with source, date, and why it matters.
- Explicit labels for confirmed facts, reasoned inferences, and remaining gaps.
Read next
docs/reference/ai-agent-web-research-stack.mdsite-extract, browser-fallback, source-audit, and compliance-and-rate-limit as needed
