| name | research-web |
| description | Standard web research workflow and source evaluation (SIFT, lateral reading, CRAAP), search strategy, triangulation, and citations. Use when asked to research topics, gather sources, or validate claims. |
Web Research
Overview
Use this skill to conduct reliable web research, evaluate sources, and produce well-attributed findings for general and technical topics (APIs, code, UI frameworks, components).
Workflow
- Define scope and recency needs (what must be current vs. historical).
- Plan a search strategy (queries, synonyms, and sources).
- Execute searches; use operators and filters when helpful (e.g.,
site:,filetype:,intitle:,inurl:). - If the topic is technical (code, APIs, UI frameworks, components), follow Technical Research Steps.
- Evaluate sources using the Source Evaluation section.
- Prefer primary sources (official docs, research papers, specs) and triangulate at least 2 independent sources for non-trivial claims.
- Record citations with dates, note conflicts or uncertainty, and summarize with clear limitations.
Search Strategy
- Use multiple query variants; include synonyms and domain-specific terms.
- Apply date filters when recency matters; record the date range.
- Prefer built-in search tools/filters for time-bounding when available.
- Use
site:to target authoritative domains, but verify currency. - Search operators and ranking behavior change over time; validate critical queries with current docs.
- For technical topics, search across layers: official docs site, source repo (issues/releases), package registry, and relevant standards/specs.
Technical Research Steps (code, APIs, UI frameworks/components)
- Identify the exact product/library name, version, and ecosystem (language, framework).
- Prefer official documentation, API references, and release notes/changelogs.
- Validate examples against the current major/minor version; note version-specific behavior.
- Check migration guides or breaking change notes when versions are close or unclear.
- Check package registries or repo releases/tags for the latest stable version and support windows.
- Review issue trackers for known breakages, regressions, or deprecations affecting the topic.
- For UI components, verify accessibility guidance, theming tokens, and required peer deps.
- If sample code is from third parties, cross-check against official docs or tests.
Topic-Specific Strategies
- Code/APIs: prioritize official API refs, type definitions, and changelogs; confirm examples compile against the stated version and note any deprecations.
- Design/Visual: look for official brand guidelines, design tokens, and component libraries; verify asset usage rights.
- UI Components: verify accessibility (ARIA/roles), theming tokens, and required peer dependencies; check component API stability.
- UX/Flows: prefer user journey specs, product requirements, and usability heuristics; distinguish behavior from visuals.
- CI/CD: find the exact CI provider docs and repo config (e.g., YAML files); confirm env vars, caches, and matrix support; verify secrets handling, fork/PR policies, and least-privilege defaults (tokens/permissions).
- GitHub Workflows: use GitHub Actions docs for syntax, contexts, and permissions; set
permissionsat workflow/job level to least privilege forGITHUB_TOKEN; prefer pinning actions to full commit SHAs; check action versions and deprecations; audit third-party action code and verify creator trust signals (e.g., verified creators). - AI agents: prioritize system/developer prompt rules, tool docs, and sandbox/permission constraints; verify behavior with current product docs and note any model/version limitations.
Documentation
- Keep a short search log: date of last search, source/platform, full query strings, filters/limits, and rough result counts.
- If you refine or rerun searches, log the updates and why.
Output Format
- Findings: concise bullet list of conclusions.
- Evidence: cite primary sources first; note version/date for technical claims.
- Uncertainties: list conflicts, missing data, or assumptions.
- Recency: call out anything that may have changed since the search date.
Source Evaluation
- SIFT: Stop before sharing, Investigate the source, Find better coverage, Trace claims to original context (go upstream).
- Lateral reading: practice click restraint, leave the page, and verify the source/claim across independent sources (do not rely only on the site’s “About” page).
- CRAAP: Currency, Relevance, Authority, Accuracy, Purpose.
Guardrails
- If sources conflict, report the disagreement and cite both.
- Avoid relying on a single source for high-stakes or fast-changing facts.
- If no reliable sources exist, say so and explain what was insufficient.
- Trace claims to primary sources when possible; if only secondary sources exist, note the limitation.
- If only one source is available, mark the claim as lower confidence.
- Watch for marketing bias or SEO content; prefer neutral or primary sources.
- If information is region- or policy-specific, state the applicable region and date.