ai-commit-msg

ai-commit-msg

Writes an AI-generated conventional commit message to the COMMIT_EDITMSG file via the prepare-commit-msg Git hook. Reads the staged diff and passes it to your configured AI provider.

Setup

1. Install and initialise husky:

npm install --save-dev husky && npx husky init

1. Create .husky/prepare-commit-msg:

#!/bin/sh
ai-commit-msg -o "$1"

1. Create a config file at ai-commit-msg.config.ts (or package.json key ai-commit-msg):

// ai-commit-msg.config.ts
export default {
  provider: "google",
  model: "gemini-2.5-flash",
};

1. Set the required environment variable:

export GOOGLE_GENERATIVE_AI_API_KEY=your-key

Core Patterns

Google provider

export default {
  provider: "google",
  model: "gemini-2.5-flash",
};

Requires GOOGLE_GENERATIVE_AI_API_KEY in the environment.

OpenAI provider

export default {
  provider: "openai",
  model: "gpt-4o-mini",
};

Requires OPENAI_API_KEY in the environment.

Ollama provider (local, no API key)

export default {
  provider: "ollama",
  model: "llama3.2",
};

Requires the Ollama server running locally (ollama serve).

Custom prompt

export default {
  provider: "google",
  model: "gemini-2.5-flash",
  prompt:
    "Write a short conventional commit message in lowercase for this diff:\n\n{{diff}}",
};

The {{diff}} placeholder is replaced with the staged diff (truncated at 8000 characters). The prompt must include {{diff}} or the model receives no diff context.

Skip the hook for one commit

export default {
  provider: "google",
  model: "gemini-2.5-flash",
  skipNextRun: true,
};

When skipNextRun is true, the hook logs a warning and exits immediately. Remove it from the config after the manual commit.

Common Mistakes

CRITICAL Default config uses ollama/llama2 — fails if Ollama not running

Wrong:

# No config file — hooks use default: provider: 'ollama', model: 'llama2'
# Works locally if ollama is running, fails in CI