claude-devkit
Health Uyari
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 5 GitHub stars
Code Basarisiz
- fs module — File system access in scripts/lib/ensure-codex-config.js
- rm -rf — Recursive force deletion command in scripts/lib/task-runner.sh
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Shared config, skills, and update scripts for Claude Code, Codex, and Gemini CLI
claude-devkit
One config. Two AI agents. Every project.
Stop copy-pasting prompt files. claude-devkit is the portable agent-config kit that keeps Claude Code and Codex aligned across every repo on your machine — shared rules, skills, slash commands, and update scripts wired in through symlinks instead of forks.
📖 From Ultrathink to Goal — A Year of AI Coding Engineering
The interactive slide deck behind this kit · 中文
Why You'll Star This
- ⚡ 30-second project onboarding — one command symlinks every rule, skill, and command into a new repo
- 🔁 Update once, propagate everywhere —
git pullhere, every project gets the new rules instantly - 🤖 Multi-agent ready — Claude Code and Codex hand off work to each other through Ralph Loop reviews
- 🧠 Real workflows, not demo prompts — distilled from daily multi-agent development, not a starter pack
- 🛠️ One updater rules them all — CLIs, MCP servers, skills, commands, and the GSD workflow stay current with a single script
30-Second Setup
# Clone the devkit once
git clone https://github.com/MiaoDX/claude-devkit.git ~/claude-devkit
# Drop it into any project
cd ~/your-project && ~/claude-devkit/scripts/setup.sh
# Install or refresh the supported tooling
~/claude-devkit/scripts/update.sh
That's it — your repo now speaks Claude Code and Codex with the same vocabulary.
| macOS — setup.sh | Ubuntu — update.sh |
Use It Everywhere
Run setup from each repo that should inherit the shared agent config:
cd ~/projects/robotics-arm && ~/claude-devkit/scripts/setup.sh
cd ~/projects/web-dashboard && ~/claude-devkit/scripts/setup.sh
cd ~/projects/ml-pipeline && ~/claude-devkit/scripts/setup.sh
Each project gets symlinks back to this devkit. Pull updates here, every project picks them up automatically.
What Gets Linked
| File or directory | Purpose |
|---|---|
CLAUDE.md |
Claude Code entrypoint — operating rules, conventions, delegation policy |
AGENTS.md |
Codex entrypoint — same intent, Codex-specific constraints |
.claude/commands/ |
Claude slash commands |
.claude/skills/ |
Claude Code skills |
context/ |
Shared reference material projects can pull from |
What's Inside
Shared Operating Rules
The same high-value behavior, every project, both agents:
- Parallel delegation for independent exploration, review, and verification
- Main-thread focus on requirements, architecture, integration, and synthesis
- Real tests over excessive mocks
- Visualization-aware validation when logs and numbers can miss rendering or geometry failures
- Repo constraints such as
fetch-mcp,uv+.venv, and conservative cleanup behavior
Multi-Agent Skills
Skills that teach Claude Code to orchestrate Codex and pair on hard problems:
| Skill | What it does |
|---|---|
| codex | Delegate analysis, review, and refactoring to Codex CLI |
| intuitive-doc | Keep AI-developed repos readable through a small human doc surface |
Ralph Loop Reviews
The Ralph Loop is an iterative review → triage → fix → verify cycle. One agent finds problems, another fixes them, both stop when convergence is reached:
| Skill | Reviews | Reviewer |
|---|---|---|
codex-plan-ralph-refactor |
GSD phase plans | Codex |
codex-impl-ralph-refactor |
Implemented code | Codex |
agent-teams-plan-ralph-refactor |
GSD phase plans | Claude agent teams |
agent-teams-impl-ralph-refactor |
Implemented code | Claude agent teams |
# Review plans before executing
/codex-plan-ralph-refactor 38
# Review code after implementation, auto-fixing must-fix issues
/codex-impl-ralph-refactor 42 --fix-level must
Slash Commands & Scripts
| Command or script | Purpose |
|---|---|
/gsd_squash |
Squash noisy commits into clean, logical git history |
/gsd_status [N] |
Show status of the last N GSD phases |
scripts/setup.sh |
Symlink this devkit into a project |
scripts/update.sh |
Install or update CLIs, MCP servers, skills, commands, and GSD |
scripts/convert-docs.sh |
Convert code/docs to LLM-ready markdown |
How It Works
Edit rules in CLAUDE.md and AGENTS.md once. Every project that has run setup.sh reads them through symlinks. No copies, no drift.
Supported Tools
| Tool | Integration |
|---|---|
| Claude Code | Primary runtime — skills, slash commands, review workflows |
| Codex CLI | Delegation skills + Ralph Loop code review |
| GSD workflow | Vendored lifecycle commands and specialist agents |
| MCP servers | fetch-mcp and related tooling installed by update.sh |
Contributing
PRs are welcome from humans and AI agents.
The most useful contributions: new skills, sharper shared rules, safer updater behavior, doc improvements, and fixes to workflows that drifted as the underlying CLIs evolved.
License
MIT — see LICENSE.
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi