Chisle
코딩 에이전트 플러그인, 20개 실시간 작업에서 기본 모델 52% 활용, 간결한 페르소나와 출력 압축 제공
하나의 .agent/ 디렉터리와 4개 에이전트 훅, MCP 서버·가드·Codex Claude형 서브에이전트 제공, 실행 시 의존성 없음
One committed .agent/ directory, hooks on four agents, an MCP server, a guard, and Claude-style subagents on Codex. Zero runtime dependencies.
개발 도구GitHub @CodeForFee온라인운영자 미확인
써 보셨어요? 한 번만 눌러 주세요. 로그인은 필요 없어요.
저장소 정보는 아직 수집하지 않았습니다.
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
One shared harness for every AI agent. Shared memory, a project map, and a handoff protocol that survives across Claude Code, Cursor, Codex and Antigravity.
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Shared project memory, task ownership and handoffs for Claude Code, Codex, Cursor and Antigravity.
Quick start · How it works · Agent support · Safety · Commands · Development
With Node.js 18 or later, run from your project:
cd your-project
npx @codeforfee/godkit init
npx @codeforfee/godkit doctor
init creates the shared project memory, installs rules, skills and role definitions, and sets up
hooks and MCP for detected hosts. Re-running it preserves project memory and user-owned settings.
Use init --no-install to configure only the project.
Then open your coding agent and work as usual. Godkit's rules guide it to read the shared state, claim files before editing and leave a verified handoff. Hooks enforce the supported parts of that workflow; coverage varies by host.
For a machine without Node.js, use the runtime-aware installer. It reuses a compatible runtime or prepares a private, checksum-verified Node 24 installation.
Codex: approve Godkit's hooks through
/hooksin the CLI or Review hooks in the app after installation and upgrades. Rungodkit doctorto inspect configuration and execution evidence.
For a global CLI, install with npm install -g @codeforfee/godkit; the examples below use godkit.
You can also use npx @codeforfee/godkit in its place.
See the installation guide for upgrades, rollback, offline setup and uninstall.
Coding agents share a filesystem, but their private session memories do not follow them between
hosts. Godkit keeps the project context in a committed .agent/ directory so the next session can
continue from recorded work.
.agent/
├── BOARD.md file claims, task index, bugs and decisions
├── THREAD.md coordination between sessions
├── BRIEF.md project goals and constraints
├── MAP.md generated codebase overview
├── graph.json machine-readable project map
├── SKILLS.md generated project skill summary
├── skills/<name>/ reusable project procedures
├── sprints/ goals split into waves of tasks
├── tasks/ plans, execution, verification and handoffs
└── log/ one handoff log per session
The protocol has two rules:
Read
.agent/before you edit. Write your log before you finish.
An agent reads the board and map, claims a file scope, does the work and records the command that verified it. Larger tasks are split by file boundaries; a sprint groups them into waves with a verification step between waves. Claims and logs name the model that performed the work.
Commit .agent/ with the changes it describes. Session logs are separate files to reduce merge
conflicts. Generated map files are regenerated rather than merged by hand.
Godkit ships skills for planning, execution, testing, review, mapping and handoffs. The rules tell the agent when to load each one; the skill catalog contains the full instructions.
godkit scan collects files and dependencies; the godkit-map skill analyzes them,
then godkit save writes the graph and overview. These commands do not analyze the code by themselves..agent/skills/<name>/SKILL.md. godkit skills
inspects them; godkit skills --link makes eligible skills available to supported hosts. Skills
declare origin and enabled; safety findings and recorded usage affect eligibility.godkit evolve summarizes evidence about project skills; godkit refactor
ranks source files by recorded churn and blame. The agent judges what needs changing.| Integration | Claude Code | Codex | Cursor | Antigravity |
|---|---|---|---|---|
| Rules | CLAUDE.md | AGENTS.md | .cursor/rules/godkit.mdc | AGENTS.md, .agents/rules/godkit.md |
| Workspace skills | .claude/skills/ | .agents/skills/ | through rules and skill paths | .agents/skills/ |
| Native role files | .claude/agents/ | .codex/agents/ | no verified format | .agents/agents/ |
| Hooks | full registration | full registration; approval required | adapted; shell-only changes untracked | adapted |
| MCP | supported | supported | supported | supported |
godkit doctor --json distinguishes configured files from features observed running in a host.
A generated role or registered hook alone does not establish that an IDE used it. Required host
trust and approval steps remain part of setup. godkit doctor --fix repairs Godkit-owned components.
See agent portability for configuration paths, hook coverage and surface-specific limitations.
여기까지가 README 앞부분입니다. 나머지는 원문에서 읽을 수 있습니다. README 전문 보기 ↗