code-agent-auto-commit

code-agent-auto-commit (cac) provides configurable code-agent-end auto-commit(using your git account) for:

• OpenCode
• Claude Code
• OpenAI Codex CLI

Features

• Auto-commit when a chat/agent turn ends
• Commit strategies:
  • single: all changed files in one commit
  • per-file: one file per commit
• AI-generated commit messages with multi-provider model configuration
  • OpenAI-compatible mode (openai-completions)
  • Anthropic-compatible mode (anthropic-messages)
• Optional auto-push to GitHub, GitLab, or generic remotes
• Tool installers for OpenCode, Codex, and Claude Code

Installation

pnpm add -g code-agent-auto-commit@latest

To update to the latest version:

pnpm update -g code-agent-auto-commit
OR:
pnpm add -g code-agent-auto-commit@latest

Then use the short command:

cac --help

Quick Start

# 1. Initialize config
cac init

# 2. Configure AI API key (pick one method):
#    Global (recommended — works across all projects):
cac ai set-key <provider|ENV_VAR> <api-key>
#    Or project-level (edit .cac/.env):
echo 'MINIMAX_API_KEY=sk-xxx' >> .cac/.env

# 3. Install hooks
cac install --tool all --scope project

# 4. Verify
cac status --scope project
cac ai "hello"

# 5. Agentic coding
opencode / claude / codex

Important: After cac init, you must configure an AI provider API key. Without a valid key, AI commit messages will not work and cac falls back to generic chore(auto): ... prefixed messages.

Model tip: Choose a fast, lightweight model (e.g. gpt-4.1-mini, MiniMax-M2.5-highspeed). Commit messages are short — speed matters more than intelligence here.

Commands

cac init [--worktree <path>] [--config <path>]
cac install [--tool all|opencode|codex|claude] [--scope project|global] [--worktree <path>] [--config <path>]
cac uninstall [--tool all|opencode|codex|claude] [--scope project|global] [--worktree <path>]
cac status [--scope project|global] [--worktree <path>] [--config <path>]
cac run [--tool opencode|codex|claude|manual] [--worktree <path>] [--config <path>] [--event-json <json>] [--event-stdin]
cac set-worktree <path> [--config <path>]
cac ai <message> [--config <path>]
cac ai set-key <provider|ENV_VAR> <api-key> [--config <path>]
cac ai get-key <provider|ENV_VAR> [--config <path>]

Command Details

• cac init: creates .cac/.code-agent-auto-commit.json under the worktree (unless --config is provided), and also writes .cac/.env.example and .cac/.env with default provider API key env vars.
• cac install: installs adapters/hooks for selected tools (opencode, codex, claude) in project or global scope. If no config exists at the resolved path, it creates one first.
• cac uninstall: removes previously installed adapters/hooks for selected tools and scope.
• cac status: prints resolved config path, worktree, commit mode, AI/push toggles, and install status of each adapter.
• cac run: executes one auto-commit pass (manual or hook-triggered). It reads config, filters changed files, stages/commits by configured mode, and optionally pushes. Hook-triggered runs also write logs to .cac/run-<timestamp>.log.
• cac set-worktree: updates only the worktree field in the resolved config file.
• cac ai: tests AI request (cac ai "hi") or manages global keys (set-key / get-key).

Config File

Default project config file:

.cac/.code-agent-auto-commit.json

Generated env templates:

• .cac/.env.example
• .cac/.env

You can copy from:

.code-agent-auto-commit.example.json

Full schema and options:

• doc/CONFIG.md
• doc/zh-CN.md

AI Key Resolution

cac resolves API keys automatically in this order (first match wins):

1. process.env — environment variable already set in the current shell
2. ~/.config/code-agent-auto-commit/keys.env — global keys file (written by cac ai set-key)
3. .cac/.env — project-level env file (created by cac init)
4. ai.providers.<name>.apiKey — key stored directly in config (not recommended)

This means hooks work out of the box — no need to manually source .cac/.env before running. If AI request fails (missing key, invalid provider/model, or non-2xx response), cac falls back to commit.fallbackPrefix-style messages.

AI Models (Multi-Provider)

Model format follows provider/model (OpenClaw-style). Example:

{
  "ai": {
    "enabled": true,
    "model": "openai/gpt-4.1-mini",
    "defaultProvider": "openai",
    "providers": {
      "openai": {
        "api": "openai-completions",
        "baseUrl": "https://api.openai.com/v1",
        "apiKeyEnv": "OPENAI_API_KEY"
      },
      "anthropic": {
        "api": "anthropic-messages",
        "baseUrl": "https://api.anthropic.com/v1",
        "apiKeyEnv": "ANTHROPIC_API_KEY"
      },
      "moonshot": {
        "api": "openai-completions",
        "baseUrl": "https://api.moonshot.ai/v1",
        "apiKeyEnv": "MOONSHOT_API_KEY"
      }
    }
  }
}

Integration Notes

• OpenCode: installs plugin under .opencode/plugins/ or ~/.config/opencode/plugins/
• Codex CLI: writes managed notify block in .codex/config.toml or ~/.codex/config.toml
• Claude Code: installs Stop hook in .claude/settings.json or ~/.claude/settings.json

If Codex config already has a custom notify = ..., installer stops and asks for manual merge.

Development

pnpm install
pnpm run typecheck
pnpm run build
pnpm test

Open Source

• Contribution guide: CONTRIBUTING.md
• Security policy: SECURITY.md
• Code of conduct: CODE_OF_CONDUCT.md

License

MIT