ai-deep-research
Installation
SKILL.md
Deep Research
Use this skill to design and run repeatable research workflows that gather evidence across many sources, preserve provenance, and synthesize results into decision-ready outputs.
This skill covers both native deep-research agents (ChatGPT Deep Research, Gemini Deep Research, Perplexity Deep Research, Claude with web search) and custom agentic research pipelines (planner / searcher / verifier / synthesizer split).
Quick Reference
| Question | Default |
|---|---|
| When to use a native agent vs custom pipeline? | Native for ad-hoc, open-ended questions. Custom for repeatable, auditable, or multi-source workflows. |
| What is the first artifact of any research task? | The source ledger — never the synthesis. |
| When is a source trustworthy? | When it is a primary document with a stable URL, author attribution, and a verifiable date. |
| What stops an unbounded research loop? | An explicit stop criterion defined before the loop starts (saturation condition or max iterations). |
| How to handle contradictory sources? | Separate them into evidence tiers; do not resolve by averaging. |
| How do I score a source? | Grade reliability and credibility separately — see references/evidence-packaging.md. |
| How do I evaluate or benchmark a research agent? | BrowseComp (arXiv 2504.12516: 1,266 hard-to-find-fact problems; Table 3 on p.6 reports an early OpenAI Deep Research at 51.5% accuracy vs 1.9% for GPT-4o with browsing, and footnote 1 notes Deep Research was trained on data that teaches BrowseComp-style tasks, a contamination caveat) and DeepResearch Bench (arXiv 2506.11763: 100 PhD-level tasks across 22 fields, scoring report quality and citation accuracy). Check the current leaderboards yourself; use ai-evals for judge reliability. |