Agent Commits

Agent Commits

A specification for AI-authored commit messages that enables human reviewers to understand why changes were made, not just what changed.

The Problem

When AI agents write code, traditional commit messages become inadequate:

  • "Add feature X" doesn't explain why the agent chose this approach
  • "Fix bug in Y" doesn't reveal the agent's understanding of the problem
  • Reviewers can't distinguish between intentional decisions and potential misunderstandings
  • No audit trail of what the human requested vs. what the agent delivered

The Solution

Agent Commits is a structured commit message format that captures:

Intent

What the human asked for

Plan

How the agent decided to accomplish it

Changes

What files were modified and why

Verification

How the changes were tested

Example

[agent] feat: Add user authentication with JWT tokens

Intent: User requested login functionality with session persistence

Plan:
- Add JWT library for token generation
- Create auth middleware for protected routes
- Add login/logout endpoints
- Store refresh tokens in httpOnly cookies

Changes:
- src/auth/jwt.ts: Token generation and validation utilities
- src/middleware/auth.ts: Route protection middleware
- src/routes/auth.ts: Login, logout, and refresh endpoints
- package.json: Added jsonwebtoken dependency

Verified: All auth tests passing, manual login flow tested

Quick Start

For Claude Code Users

Add the Agent Commits skill to generate compliant commit messages automatically:

/add-skill https://agentcommits.com/skill.md

Manual Usage

Follow the specification to format your AI-assisted commits, or check out real-world examples.

Why This Matters

For Reviewers

  • • Understand the original request
  • • Verify plan aligns with intent
  • • Spot misunderstandings early
  • • Trust but verify AI changes

For Teams

  • • Maintain audit trails
  • • Clear accountability
  • • Enable better tooling
  • • Build institutional knowledge

For Agents

  • • Structured format reduces ambiguity
  • • Clear expectations
  • • Enables self-verification
  • • Supports iteration

Get Started

Read the full specification or start using Agent Commits today.