Research

Research

Use this skill to turn a broad question into a source-grounded answer with explicit uncertainty.

Operating Rules

• Use current sources when facts may have changed.
• Prefer primary sources: official docs, standards, papers, filings, source repositories, release notes, and direct data.
• Verify URLs before presenting them.
• Separate sourced facts from inference.
• Do not invent citations, quotes, dates, statistics, or source names.
• Keep scope proportional: use the smallest research pass that can answer the question well.

Mode Selection

• Quick research: 1-3 high-quality sources; use for narrow factual checks or lightweight comparisons.
• Standard research: 4-8 sources; use for recommendations, tradeoff analysis, market scans, or implementation choices.
• Deep investigation: iterative source gathering; use for ambiguous claims, high-stakes decisions, conflicting evidence, or complex landscapes.
• Technical research: prioritize official documentation, changelogs, source code, specs, and reproducible examples.
• Literature research: prioritize papers, preprints, citations, author/institution context, and methodology limits.

Workflow

1. Define the exact question, decision, and desired output.
2. Identify source classes needed to answer it.
3. Gather sources and discard weak or duplicative material.
4. Cross-check important claims against independent evidence.
5. Capture uncertainty, caveats, and known disagreement.
6. Synthesize into a concise answer with links to the sources used.

Source Quality Heuristics

• High confidence: primary source, recent enough for the question, directly supports the claim.
• Medium confidence: reputable secondary source, consistent with primary evidence, but indirect.
• Low confidence: unsourced summary, unclear provenance, marketing copy, social post, or stale page.

Output Pattern

For most research tasks, return:

• Answer: direct response or recommendation.
• Evidence: key source-backed findings.
• Tradeoffs: what changes the answer.
• Uncertainty: unresolved questions or weak evidence.
• Sources: links used, grouped by relevance.

For decisions, end with concrete next steps.