kimi_cli.md

Kimi CLI

Kimi CLI does not have lifecycle hooks yet — it's an open feature request. Enforcement is advisory only, through AGENTS.md and MCP config. github

1. Add an AGENTS.md file

Kimi CLI reads AGENTS.md from the project root (you can generate one with /init). Add a git policy section: stevekinney

# AGENTS.md

## Git Policy
- Never run `git commit`, `git push`, or `gh pr create` directly.
- For commits, use `omni commit --yes --json`.
- For pull requests, use `omni pr --yes --json`.
- Always pass `--json` for structured output and `--yes` for non-interactive mode.

## Omni CLI Reference

| Task | Command |
| --- | --- |
| Quick commit | `omni commit` |
| Preview commit | `omni commit --preview --json` |
| Apply commit | `omni commit --apply --json` |
| Preview PR | `omni pr --preview --json` |
| Create PR | `omni pr --apply --json` |
| CI commit | `omni commit --yes --json` |
| CI PR | `omni pr --yes --json` |

2. (Optional) Add Omni as an MCP tool

If Omni exposes an MCP server, you can register it in your MCP config so Kimi discovers it as a first-class tool: blog.gitbutler

{
  "mcpServers": {
    "omni": {
      "command": "omni",
      "args": ["mcp-server"]
    }
  }
}

Then run Kimi with:

kimi --mcp-config-file /path/to/mcp.json

This makes omni available as a tool Kimi can call directly, rather than shelling out via run_shell_command. blog.gitbutler

Limitations: Without hooks, there is no way to hard-block git commit at the system level. The agent should follow AGENTS.md instructions, but it's not guaranteed. Once the hooks feature request lands, you'll be able to intercept shell commands the same way Claude Code and Cursor do. github