Research

Use this researcher-owned skill to research source material and produce a standalone research dossier before generating official skill eval cases.

Work primary-source-first. Resolve the task boundary and missing constraints before searching. End with approved sources and mapping notes, not guessed eval rows.

When to use this skill

• The optimizer or evaluation workflow needs grounded eval cases, but no suitable local source material exists yet.
• The user wants grounded public examples instead of purely simulated rows.
• The agent needs a ranked shortlist of public datasets, benchmarks, or documentation sources that match a prompt task.
• The user needs explicit judgment about source quality, data reliability, annotation quality, licensing, provenance, or leakage risk before authoring eval data.
• The workflow needs to know whether no acceptable public source exists, so synthesis should stop instead of guessing.

If the source material is already known and the job is to convert it into eval rows, use a synthesis workflow instead.

Inputs