research-public-weblisted

# Research Public Web Answer a research question using **public, no-key or local adapters** first. Paid SERP/search APIs are optional fallbacks, not the default. ## When to use - Competitive/market/docs research on public pages - Need citeable URLs and excerpts without SerpAPI budget - Validating claims against Wikipedia, package registries, or public HTML ## When not to use - Private corp intranet (use authenticated tools) - Paywalled content you are not licensed to fetch - Deep browser automation for authenticated app flows (different job) ## Method **Open [references/recipes.md](references/recipes.md) first** — copy-paste endpoints/commands; do not web-search for these defaults. ### 1. Frame the question - Write the decision the research must support. - List must-have sources (official docs, registry, standards) vs nice-to-have. ### 2. Search with free paths Default ladder: 1. **Host web tools** already available (browser/search tools on the agent host). 2. **Public HTML search adapters** (DuckDuckGo HTML class) and vertical indexes (HN Algolia, Wikipedia API). 3. **Direct official URLs** when known (docs, RFCs, registry package pages). 4. **Archive.org** when live page is gone or changed. 5. Paid SERP only if free paths fail and budget exists—label results L4. Open [references/providers/INDEX.md](references/providers/INDEX.md). ### 3. Fetch and extract - Prefer primary sources over blog SEO mirrors. - Capture URL, retrieved time, and short excerpt with loca