Installation
You need a Codex environment that supports local plugins and Python 3.10 or later. Download the full repository, including scripts/, and run these commands from the root containing plugin.json.
mkdir -p .agents/plugins
cp -n marketplace.example.json .agents/plugins/marketplace.json
codex plugin marketplace add .
codex plugin add research-compass@research-tools
If you already use .agents/plugins/marketplace.json, add the example plugin entry to its plugins array. After installation, start a new chat in the Codex app.
Usage
A topic is enough to start. Include data access, research duration, hardware, or preferred journals and conferences to narrow the recommendations.
$research-compass:research-compass
Explore pre-training research topics in medical imaging AI.
I have public datasets, two months, and one RTX 4070 SUPER GPU.
Compare each candidate with prior work and propose a first-week pilot.
Name the journals and conferences that fit, explain why, and save the results in a session folder.
Full-text review and venue verification require access to the source papers and official pages. Unavailable information is marked as unverified.
Output files
| File | Contents |
|---|---|
report.md | Candidate comparison, research plans, and submission strategy |
evidence-cards.md | Source locations, paper versions, and excerpts for claims |
first-week.md | First-week validation checklist |
references.bib, references.ris | Bibliography exports |
state.json | Research constraints, search history, and progress |
To resume, invoke the skill with the path to state.json and describe the changes you want. If you upload files directly to GitHub, exclude runs/, which contains private research results.
The workflow guide covers pilot measurements, search budgets, venue rechecks, and final report validation. The data contract covers CLI usage and file formats. Both guides are in Korean.
License
Research Compass is released under the MIT License.