Deep Research

What it does

• Decomposes a research question into 4–6 (or 3–4 in Lightweight mode) specialist tasks, dispatches subagents in parallel, and keeps raw search results out of the lead agent's context (~60–70% context savings).
• Classifies every source by type (official, academic, secondary-industry, journalism, community) and accessibility, then enforces quality gates: ≥30% official sources, ≥5 unique domains, ≤25% single-source share.
• Builds a locked citation registry — after P3 no new sources may appear and dropped sources can never be resurrected. Every claim carries an [n] citation plus a High/Medium/Low confidence marker.
• Runs a mandatory counter-review (P6) that must surface at least 3 weaknesses — single-source claims, stale data, contradictions — and publishes them as a "Key Controversies" section.
• Enterprise mode adds six-dimension data collection, SWOT, weighted competitive-barrier scoring, a risk matrix, and a scorecard across a 7-chapter template.

Who it's for

• Analysts, consultants, and PMs producing market, industry, or competitive reports
• Researchers and grad students who need traceable citations, not vibes
• Anyone worried about AI-hallucinated statistics who wants verification baked into the workflow

Example uses

1. "Research the competitive landscape for AI code review tools" → 5 parallel tasks → citation registry → 5,000-word report with confidence markers and controversies.
2. "Do a due-diligence brief on Company X" → Enterprise mode collects six dimensions and outputs a 7-chapter report with SWOT and a probability × impact risk matrix.
3. "Compare three vector databases for our stack" → Lightweight mode, 3–4 tasks, a 2,000–4,000-word technical brief with sourced trade-offs.

Install the skill I found on Claude Skill Mart.
Copy the deep-research folder from the GitHub repo daymade/claude-code-skills into my ~/.claude/skills/deep-research/.
When it's done, tell me in one line what this skill can do.