Kim Orchestrator workflow
Ten Claude Code commands that run Codex and Gemini as helpers: Claude plans, Codex writes code, Gemini reviews and reads mockups
- 119 ★ on the repository
- Last commit
Extended access
Why this level
- The commands send code and requirements to external OpenAI and Google models and spend money on their accounts
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/kim-orchestratorDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The commands send code and requirements to external OpenAI and Google models and spend money on their accounts
This is third-party code. Review the repository files before installing.
What it does
Kim Orchestrator tackles the problem of scattered AI tools: Claude's output used to be copied by hand into Codex and then into Gemini. Here one command starts a chain in which Claude analyses requirements and coordinates, Codex generates code and Gemini reviews it. The author reports a time saving from 20 minutes to 7 on his own example. The `.claude/commands` folder holds ten commands in order of use: `/kim-setup` checks the environment, `/kim-help` suggests a command, `/kim-plan` breaks requirements into tasks, `/kim-crud`, `/kim-api` and `/kim-form` generate CRUD, endpoints and forms through Codex, `/kim-code` writes code without review, `/kim-review` hands code to Gemini for review, `/kim-team` runs all three engines and `/kim-ui2code` turns a mockup screenshot into a frontend. The repository also has a `kim-orchestrator` skill with prompts for three phases, MCP wrappers for Codex and Gemini, a JWT auth example on FastAPI and Chinese documentation.
Who it is for. Developers who already use Claude Code, Codex and Gemini CLI and want to pass work between them with one command instead of copying by hand.
Good fit when
- A big feature must be split into tasks with `/kim-plan` and then generated through Codex
- A users table has to become CRUD or an endpoint with `/kim-crud` and `/kim-api`
- A finished file is checked by a second model: `/kim-review src/auth/utils.py`
- A login-page mockup screenshot becomes a frontend with `/kim-ui2code`
Not a fit when
- You have no access to Codex or Gemini CLI: the chain relies on both
- The task is small: the author himself says simple questions need plain dialogue, no commands
- You need English-language support: the docs and command interface are written in Chinese
Example request
/kim-team implement JWT login: registration, sign-in, token refresh.Limitations
You must install three CLIs: Claude Code, Codex and Gemini CLI, and Gemini needs browser sign-in. The chain spends money with three providers, and the README prices are from December 2025 and may have changed. All commands and docs are in Chinese, the README changelog was last updated on 8 December 2025 and the last commit is from 16 October 2026. The JWT sample on FastAPI is a demo and not ready for production without review. The time and price gains are the author's own example.
How to disable. Delete the .claude/commands folder and the kim-orchestrator skill from the project and remove the Codex and Gemini MCP entries from .mcp.json. Remove the global CLIs only if you no longer need them.
FAQ
What do I need to install?
Three CLIs: `npm install -g @anthropic-ai/claude-code`, `npm i -g @openai/codex` and `npm install -g @google/gemini-cli`. For Gemini run `gemini auth login`, which gives 1,000 free requests a day.
Which command fits a large task?
`/kim-plan` handles planning and `/kim-team` runs the full production-grade cycle across all three engines. For small code `/kim-code` is faster because it skips review.
Who does what?
Claude analyses requirements and coordinates, Codex generates code, Gemini reviews and reads images. The author justifies the split by each model's price and quality.
Which integration methods are shown?
Five: commands, a skill, MCP, subagents and hooks. Commands are the easiest start, the skill and MCP suit long collaboration, and hooks suit automatic checks.
How do I turn a mockup into code?
Run `/kim-ui2code "./designs/login-page.png"`. Gemini reads the image, Claude plans, Codex writes the code and Gemini reviews the result.
Discussion
No questions or reviews yet.
Related
WorkflowsA workflow template for Claude Code and five more agents: a backlog task, a diff review and a commit.
A Claude Code plugin: 39 commands for PRs, review, commits and specs, nine expert roles and hooks, in eight language versions
Kimi Bridge for Claude Code
kimi-bridge
An MCP server that turns a local Kimi Code CLI into a second agent inside Claude Code, with a dedicated BSL review tool for 1C
kimaki turns Discord into a control surface for OpenCode: every channel is a project, every thread a separate coding-agent session.
Shared Claude Code settings for a team: hooks against dangerous commands, a check before finishing and skills from spec to review
A Claude Code skill that drives Codex over JSON-RPC: tasks, code reviews, parallel research and messaging between two agents.
README in short
The README is in Chinese with a December 2025 changelog. It explains the problem of disconnected AI tools, compares Claude, Codex and Gemini prices, lists five integration methods, gives installation of the three CLIs, a table of ten commands, examples, a command decision tree and the project layout with a FastAPI JWT example.