Skip to content

jeremyknows/deep-research

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Deep Research 🔬

Status: GA — A structured 7-stage research protocol for questions that require more than a quick lookup. Goes from question to defensible brief with forced hypothesis generation and adversarial challenge before writing conclusions.


What it does

  • Scopes research questions before gathering — prevents chasing the wrong thing
  • Forces hypothesis commitment before evidence gathering (prevents confirmation bias)
  • Builds an argument skeleton before filling in sources
  • Adversarially challenges the hypothesis with a steelmanned counterargument (Stage 6)
  • Produces structured briefs with calibrated confidence levels
  • Routes output correctly: inline for short answers, filed to docs/research/ for long-form

Install

Claude Code / Cursor / any Agent Skills-compatible agent:

git clone https://github.com/jeremyknows/deep-research ~/.claude/skills/deep-research

OpenClaw:

git clone https://github.com/jeremyknows/deep-research ~/.openclaw/skills/deep-research

Usage

Natural language triggers

"Do a deep research on X"
"/deep-research [question]"
"Research this properly — I need to make a decision"
"Do a structured brief on [topic]"
"Deep dive on [tool/technology/concept]"

When to use this vs other research tools

Tool Best for
deep-research Tool eval, competitive intel, concept validation, decisions you'll act on
last30days Recent news, trends, X discourse (≤30 days)
Quick web search Factual lookups, "what's the current version of X"

The 7 Stages

Stage 1: Scope       Define the question and success criteria
Stage 2: Gather      Cast wide — web, X, GitHub, docs
Stage 3: Hypothesize What's the non-obvious answer?
Stage 4: Skeleton    Build the argument structure first
Stage 5: Evidence    Fill the skeleton with sources
Stage 6: Challenge   Steelman the counterargument
Stage 7: Output      Write the brief with confidence level

Fast-path (1→2→7): Only when the question has one clear answer, stakes are low, you already have 2+ primary sources, and no significant decision depends on it.


Output format

# Research Brief: [Question]
**Date:** YYYY-MM-DD
**Confidence:** High / Medium / Low — [one sentence why]

## Answer
[1–3 sentence direct answer. Lead with the conclusion.]

## Evidence Summary
- [Key finding 1 — source]
- [Key finding 2 — source]

## Strongest Counterargument
[From Stage 6]

## Decision Recommendation
[Watch / Use now / Act on / Archive — with specific next step]

Confidence levels

Level Criteria
High 3+ independent primary sources + weak counterargument + no major unknowns
Medium 1–2 primary sources, OR counterargument has merit, OR significant unknowns
Low Mostly secondhand/opinion, OR strong counterargument, OR major unknowns

Reference files

File Purpose
references/source-routing.md Which tools/sources to use by question type
references/brief-quality-checklist.md Pre-filing checklist for briefs
references/skeleton-examples.md Stage 4 skeleton examples + Stage 6 worked examples

Health

  • Version: 1.2.0
  • Health score: 12/14
  • Status: GA — 4 production runs completed 2026-03-23, avg 5.1/6 scorecard
  • PRISM reviewed: 2026-03-22 (6 reviewers, all AWC)
  • Category: Research & Analysis

File structure

deep-research/
├── SKILL.md                              — Core skill (282 lines)
├── README.md                             — This file
├── LICENSE.txt                           — MIT
├── .gitignore
└── references/
    ├── source-routing.md                 — Source selection by question type
    ├── brief-quality-checklist.md        — Pre-filing quality checks
    └── skeleton-examples.md              — Stage 4/6 worked examples

MIT License — jeremyknows — v1.2.0

About

7-stage structured research protocol for defensible briefs

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors