magi-researchers

Category: 数据与AI | Uploader: Axect | Downloads: 0 | Version: v1.0（Latest）

Run the MAGI multi-model research workflow from Codex. Use when the user wants brainstorming, planning, implementation, testing, reporting, or concept explanation through coordinated Claude, Gemini, and Codex passes, especially as a user-level reusable workflow outside the original repository.

Changelog: Source: GitHub https://github.com/Axect/magi-researchers

Directory Structure

Current level: .codex-skill/magi-researchers/

📁 agents/
- 📄 openai.yaml 162 B
📁 references/
- 📄 local-install.md 159 B
📄 AGENTS.md 2.8 KB
📄 SKILL.md 2.8 KB

SKILL.md

---
name: magi-researchers
description: Run the MAGI multi-model research workflow from Codex. Use when the user wants brainstorming, planning, implementation, testing, reporting, or concept explanation through coordinated Claude, Gemini, and Codex passes, especially as a user-level reusable workflow outside the original repository.
---

# MAGI Researchers

Use this skill when the user wants the MAGI workflow from Codex in any workspace, not only inside the original `magi-researchers` repository.

## Quick Start

1. Read `references/local-install.md` to confirm the installed skill path and bundled resources.
2. If external tools are not configured yet, run `scripts/setup_mcp.sh`.
3. Map the request to one workflow:
   - Full pipeline: `references/workflows/research.md`
   - Brainstorm only: `references/workflows/research-brainstorm.md`
   - Explain only: `references/workflows/research-explain.md`
   - Implement only: `references/workflows/research-implement.md`
   - Test only: `references/workflows/research-test.md`
   - Report only: `references/workflows/research-report.md`
4. Execute that workflow in the user's current workspace unless the user explicitly wants another location.

## Execution Rules

1. Treat the copied workflow files in `references/workflows/` as the source of truth.
2. Keep generated artifacts in the active workspace by default:
   - `outputs/...`
   - `src/`
   - `tests/`
   - `plots/`
3. Use the bundled domain templates in `references/templates/domains/` when the workflow calls for them.
4. Prefer Codex MCP tools for Codex and Gemini calls when available.
5. If a workflow step needs Claude and no Claude MCP tool exists, call `scripts/run_claude_print.sh`.
6. Preserve the artifact names defined by the workflow. Do not rename files unless the user explicitly asks.

## Slash Command Mapping

- `/magi-researchers:research ...` -> `references/workflows/research.md`
- `/magi-researchers:research-brainstorm ...` -> `references/workflows/research-brainstorm.md`
- `/magi-researchers:research-explain ...` -> `references/workflows/research-explain.md`
- `/magi-researchers:research-implement ...` -> `references/workflows/research-implement.md`
- `/magi-researchers:research-test ...` -> `references/workflows/research-test.md`
- `/magi-researchers:research-report ...` -> `references/workflows/research-report.md`

## Claude Interop

Use:

```bash
scripts/run_claude_print.sh "Review outputs/topic/plan/research_plan.md for fatal flaws."
```

If context is large, write the prompt to a file and call:

```bash
scripts/run_claude_print.sh --prompt-file /tmp/magi_prompt.txt
```

## Notes

- This installed skill is a Codex-facing distribution layer. It does not replace the original Claude Code plugin.
- When the original repository is available, prefer its files for updates. The installed skill is a snapshot created by the installer.

Comments 0

Latest | Hot

Please login before commenting.

No comments yet. Be the first one!

magi-researchers

Directory Structure

SKILL.md

Report

Notice