Deep Research — Disciplined Meta-Research
Deep Research is a disciplined, multi-source research skill that turns a high-stakes question into an auditable investigation. It fans out web searches across many channels, uses parallel sub-agents, triangulates every claim with at least three independent sources, runs an adversarial review, and saves each source to its own file with verbatim quotes. The output is a reusable folder containing a plan, source files, findings, and a final report.
✨ What it does
- Falsifiable hypotheses are defined up front and explicitly confirmed or refuted.
- Parallel sub-agents fan out across many channels for efficient searching.
- Triangulation requires at least 3 independent sources of different types per thesis.
- Each source is saved to its own file with metadata and verbatim quotes.
- An adversarial pass includes self-critique and active search for counter-arguments.
- A refresh protocol generates refresh_targets.md for delta updates later.
🎯 When to use it
- When a low-quality answer is expensive, such as for strategy, business plans, reports, or article groundwork.
- When comparing N institutions, products, methodologies, or markets and you need defensible reasoning.
- When validating a hypothesis or a decision against external data.
- When doing meta-research to understand how a field works or map a landscape.
🚀 How to use
Trigger deep-research by asking for a thorough, multi-source investigation on a high-stakes question. Provide the research question, the underlying decision, and any context. The skill will run through its pipeline and produce a folder with plan, sources, findings, and report. Example prompts:
Run deep research on the future of solid-state batteries for our strategic roadmap.
I need a deep-research investigation comparing remote work policies across top tech companies.
📄 Output: A folder containing plan.md, sources.csv, per-source files, findings, refresh_targets.md, and a dated final report.
📦 Add this skill to Claude Code
# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills
# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/research/deep-research/skills/deep-research .claude/skills/deep-researchSkill source: research/deep-research/skills/deep-research/SKILL.md
⚠️ Good to know
This is a heavyweight process not suited for quick fact-checks or low-risk overviews; it requires sufficient API keys/channels and may fall back to HTML where needed.
❓ FAQ
How does deep-research ensure claims are reliable?
It requires every thesis to be backed by at least three independent sources of different types, and each source is saved with verbatim quotes so claims can be traced.
Can I update the research later without redoing everything?
Yes, the skill generates refresh_targets.md, and running an update produces a delta report instead of replaying the whole investigation.
What happens if a source fetch returns nothing?
The claim is left empty; the skill never fabricates citations, so you won't see invented URLs.
🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.