Agent Commits That Follow Repo Conventions

Where this fits

You're halfway through a feature branch. Your agent has edited six files, run the tests, and now needs to land a clean commit. Without guardrails, it will stage everything, guess at a message, and trip over pre-commit hooks. The commit skill gives the agent a narrow, explicit routine: inspect the diff, stage deliberately, and write a message that matches the project's conventions.

Why agents benefit

• Inspects git status and git diff before staging, so it commits only intended changes—not stray debug files.
• Resolves repo-specific config (.git-workflow/config.yaml or .claude/config.yaml) to learn commit format rules before writing the message.
• Handles pre-commit hook failures gracefully, reading the error and revising the commit instead of abandoning it.
• Uses AskUserQuestion when a commit needs human judgment, like selecting between breaking and non-breaking change types.
• Respects the project's own tooling: if the repo uses a commitlint config, the agent works within it rather than imposing an external style.

Practical scenarios

Mid-feature checkpoint commit

The agent finishes a working increment across multiple files. It reviews the diff, separates the feature change from an unrelated whitespace fix, and creates a focused commit with a conventional message—without being asked to "just commit everything."

Fixing a pre-commit failure

A hook rejects the commit because of lint or formatting. The agent reads the failure output, runs the formatter on the staged files, re-stages, and retries. The developer never sees a broken tree.

PR context from existing metadata

In a repo with a PR open, the skill loads .git-workflow/pr-context.json and uses branch, issue, and PR numbers to prefix the commit message correctly—matching conventions that a generic "write a good message" prompt would miss.

Add it to your agent workflow

Install the skill into your agent runtime:

npx skills add rlajous/claude-code-commands --skill commit

The skill exposes commit for direct invocation and can also run automatically because disable-model-invocation is false. After installation, a developer simply says: "commit these changes following repo conventions"—the agent reads config, stages deliberately, and produces a formatted commit without manual review of each flag.

Compare before adopting

Before adopting, evaluate the skill's freshness: commits in the last 12 months, open issues, and whether it supports your git host's quirks. Compare it against generic Git automation scripts and your agent's built-in commit capabilities. The strongest signal is whether it matches how your team already structures commits—conventions differ wildly across repos.

Why it is worth tracking

At 30 stars, this is an early-stage skill, but it fills a real gap: agents are good at writing code and bad at respecting repository etiquette. The maintainer keeps it runtime-agnostic, which matters for portability across agent environments. Track it if your team relies on conventional commits or strict hook policies—and revisit when the project gains more adoption and battle-tested scenarios. See the latest updates and community notes at https://www.openagentskill.com/skills/rlajous-commit when evaluating fit.