199-biotechnologies/claude-deep-research-skill
Overview
This skill conducts enterprise-grade deep research using an 8-phase pipeline to deliver citation-backed, verifiable research reports. It emphasizes source credibility scoring, parallel retrieval, and automated verification to minimize fabrication and maximize evidence quality. Use it when you need a thoroughly validated, multi-source synthesis rather than a quick lookup.
How this skill works
The skill decomposes requests into multiple search angles, launches parallel retrieval agents, and triangulates claims across 10+ sources. It enforces immediate citation for factual claims, runs automated citation and structure validators, and assembles reports progressively into Markdown, McKinsey-style HTML, and PDF formats. Final outputs include an executive summary, evidence table, source quality assessment, and organized folder with all files.
When to use it
- Complex research requiring 10+ sources and rigorous verification
- Comparative analyses ("Compare X vs Y") with balanced evidence
- Decision-support reports for leadership or procurement
- Trend analysis requiring synthesis of recent data (last 1–2 years by default)
- Critical or high-risk choices needing red-team critique and validation
Best practices
- Choose mode explicitly when stakes are high: standard, deep, or ultradeep
- Provide a clear research question and any required date range or scope constraints
- Expect progressive delivery: summary first, then detailed sections written to file
- Request explicit citations for any claim you plan to act on or publish
- Allow time for verification gates (citation validation and report validation)
Example use cases
- Deliver a 4,000+ word competitive landscape with cited market numbers and vendor comparisons
- Produce a decision memo comparing architectural approaches with pros/cons and evidence
- Generate a trend report on AI safety developments, with source credibility scoring
- Create a reproducible research folder containing MD, HTML (McKinsey-style), and PDF versions with openable outputs
FAQ
Standard mode typically runs 5–10 minutes and targets 15–30 sources; deep modes take longer (10–45 minutes) depending on scope.
Will the skill fabricate citations?
No. The skill enforces an anti-hallucination protocol: every factual claim must cite a verified source and automated checks reject or flag unverified citations.
