omni-research

Omni-Research

Autonomous research agent skill for Claude Code. Give it a topic, and it autonomously searches the web, verifies sources, synthesizes findings, and produces a structured research report — all running in the background while you continue working.

/omni-research

Why This Exists

Asking an AI to "research X" gives you a one-shot answer based on training data. Omni-Research is different: it runs a multi-cycle autonomous loop that searches the web, reads actual pages, verifies every claim at the source, and builds a growing knowledge base over 5-30+ research cycles. The result is closer to what a human research analyst would produce — with citations you can actually check.

How It Works

You                          Claude Code                        Background Agent
 |                               |                                    |
 |  /omni-research               |                                    |
 |------------------------------>|                                    |
 |  "Research X topic"           |                                    |
 |                               |  Interactive setup:                |
 |  <-- Topic? Type? Scope? -->  |  1. Refine topic                   |
 |  <-- Research lines? -------> |  2. Classify type                  |
 |  <-- Confirm? --------------> |  3. Propose research lines         |
 |                               |  4. Configure depth                |
 |                               |                                    |
 |                               |  Launch background agent --------->|
 |                               |                                    |
 |  "Agent launched. Check       |    PHASE 0: Source Scouting        |
 |   research.md anytime."       |    - Hunt curated lists, blogs,    |
 |                               |      community threads             |
 |  (continue your work)         |    - Score sources (1-5)           |
 |                               |    - Build Source Queue (8+ items) |
 |                               |                                    |
 |  (edit steer.md to redirect)  |    PHASE 1: Seed Mining            |
 |          |                    |    - WebFetch & deep-read sources  |
 |          v                    |    - Extract findings with inline  |
 |   "Focus on Line 3"          |      citations                     |
 |                               |                                    |
 |                               |    PHASE 2: Hypothesis Loop        |
 |                               |    1. Check steer.md for redirects |
 |                               |    2. Pick biggest knowledge gap   |
 |                               |    3. WebSearch for evidence       |
 |                               |    4. WebFetch & verify sources    |
 |                               |    5. Synthesize with inline cites |
 |                               |    6. Update research.md           |
 |                               |    7. Every 4th: PIVOT/REFINE gate|
 |                               |    8. Every 5th: discover sources  |
 |                               |    9. Repeat until saturated       |
 |                               |                                    |
 |                               |    COMPLETION:                     |
 |                               |    - Anti-fabrication verify pass  |
 |  <-- "Research complete!" ----|--- Generate BRIEF.md --------------|
 |                               |                                    |

Features