ymping666 / CodexLab

# CodexLab
**A Multi-Agent Research Lab for Codex.**
**Already have a Codex plan? Install one Skill and start a research lab in your current project—without another LLM API key.**
Many API-driven multi-agent frameworks ask you to configure model API credentials, pay for separate API calls, and integrate another orchestration runtime. CodexLab uses your signed-in Codex environment and its native subagents. Install the Skill, invoke `$codexlab` with your research direction, and keep working in the project you already opened.
Use your eligible existing Codex plan; normal quotas apply. The default path requires no ZIP download, workspace switch, Python installation, or terminal commands from you.
GitHub · 简体中文 · Installation guide · Skill instructions · Roadmap
*v0.2 alpha · Actual bundled team-menu screenshot. The menu defaults to English; switch to 中文 without losing your choices. Interactive presentation requires a host with inline HTML support; other clients receive a text menu. The optional artifact dashboard is a separate read-only tool.*
## Install, then invoke
In Codex chat, ask the official `skill-installer` to install this repository's self-contained Skill:
 [code] 
After installation completes, try it in your **next message**:
 [code] 
Choose a purpose, review the proposed team, and adjust individual responsibilities. Confirm in chat, then describe your topic and authorized task. You can also supply a research task directly and skip team setup. If the client has not discovered the Skill, refresh its Skill list or follow its reload/restart guidance, then try again. The official installer supports GitHub repository paths; installation is complete only after your own Codex confirms it.
CodexLab writes research artifacts under:
 [code] 
The invocation session acts as PI and uses four focused worker roles as needed. You stay in the current project; the default Skill workflow does not require opening a separate lab workspace.
## Choose a purpose, then a team
The guided menu starts with four concrete uses: **read papers and find a direction**, **refine an idea**, **design or advance experiments**, and **check results or reproduction**. Each opens a small suggested team with responsibilities shown beside names. Adjust one member at a time; all styles and five complete presets remain available on demand.
| Responsibility | Three working styles |
|---|---|
| PI / current chat | Aster · synthesis / Orion · exploration / Quinn · decision focus |
| Literature | Atlas · systematic mapping / Scout · frontier scan / Flint · counterevidence |
| Method | Nova · mechanisms / Theo · theoretical constraints / Mira · minimal empirical test |
| Experiment | Forge · engineering reproduction / Vector · statistical robustness / Pulse · small pilots |
| Independent review | Sage · comprehensive audit / Rook · adversarial critique / Trace · reproduction audit |
These are fifteen **instruction profiles across five responsibilities**, not fifteen concurrent agents or di