research-answer

Use a research-then-answer workflow that searches authoritative sources before responding and attaches an inline citation to every factual claim. This Skill chains search queries, reads sources directly, and wraps the final synthesized answer with source references.

Run the claim through a structured research workflow: search for the topic, visit authoritative sources directly, extract facts with attribution, and cross-reference multiple sources. Contradictions are flagged rather than silently resolved.

Search before answering whenever information is time-sensitive, such as current events, recent changes, or fast-moving topics. Training data alone is unreliable for anything that may have changed since the model's knowledge cutoff.

Simple facts need one search and a direct answer, while deep research uses multi-query chains where each finding informs the next query. Deep research also cross-references sources, evaluates evidence quality, and distinguishes confirmed facts from speculation.

Key risks include fabricated citations, overconfidence from a single source, and confirmation bias toward agreeing sources. This workflow mitigates them with sanity checks, but very large research tasks may exceed token budgets and require continuation in a follow-up session.