RabbitHolify

RabbitHolify, formerly Deep Research Loop, turns any topic, URL, repository, person, paper, podcast, company, or concept into an interactive, source-backed research chain.

Each loop preserves findings, citations, verbatim excerpts, precise locators, contradictions, dead ends, and open branches. Continue with a single number, branch into a promising thread, synthesize across loops, export the chain, or stop with a final research report.

Included skills

Skill	Purpose
rabbitholify	Run the research loop directly.
load-rabbitholify	Activate the root skill from this repository.

Install

Copy this repository to your Codex skills directory:

git clone --branch v1.1.1 --depth 1 https://github.com/Alt3r3dP3rc3ption/RabbitHolify.git ~/.codex/skills/rabbitholify

Restart Codex, then invoke the skill:

Use $rabbitholify to research:

TOPIC SEED: <topic or URL>
OPTIONAL CONTEXT: <what you already know>
OPTIONAL FOCUS: <the angle to prioritize>

Natural-language requests work too:

Use $rabbitholify to investigate the architecture and failure modes of local-first software.

For traceable document research:

Use $rabbitholify in verbatim evidence mode on these contracts, specifications,
and drawings. Give every excerpt a quote ID, reproduce it exactly, and cite the
file, revision, page or sheet, clause/detail, and a clickable source link.

Loop controls

Choice	Action
1	Go deeper
2	Follow the highest-ranked branch
3	Synthesize completed loops
4	Start a new seed
5	Show the loop map
6	Export the research chain
7	Give a custom direction
0	Stop with a final synthesis

Design

• Source-backed: factual findings cite accessible evidence.
• Verbatim when requested: exact excerpts remain separate from analysis and carry source links, version identifiers, and page/clause/sheet-level locators.
• Local-first: authorized personal files can be quoted exactly without uploading them; third-party exports can use stable source IDs instead of exposing private paths.
• Stateful: later loops build on earlier findings without repeating them.
• Honest about limits: inaccessible sources, rejected evidence, contradictions, and bounded graph traversals stay visible.
• Self-terminating: choice 0 closes the chain cleanly.

License

Apache-2.0