Aider as a pre-commit hook: AI fixes your lint before you push

This recipe makes Aider the last line of defense between sloppy code and your remote. Lint fail? Aider fixes it. Type error? Aider fixes it. Test fail? Aider tries to fix it (and asks if it can’t).

Setup

pip install aider-chat pre-commit

Add .pre-commit-config.yaml to your repo:

repos:
  - repo: local
    hooks:
      - id: aider-fix-lint
        name: aider auto-fix lint
        entry: bash -c 'ruff check . --fix --exit-zero || aider --no-stream --message "Fix the remaining lint issues. Run ruff check . to verify."'
        language: system
        pass_filenames: false
        always_run: false
        stages: [commit]

Install:

pre-commit install

What happens on commit

1. ruff check . --fix runs first (cheap, fixes 95% of issues).
2. If anything’s left, aider is invoked non-interactively with a fix-it prompt.
3. Aider edits files, makes a commit, you re-run the hook → green.

Caveats

• Cost: every commit could be a few API calls. For a busy repo this adds up. Set a daily budget alert in your provider dashboard.
• Loops: if Aider can’t fix the issue, the commit fails. That’s correct behavior — don’t --no-verify your way past it.
• Reviewability: Aider commits with its own message. Configure a prefix like chore(auto): so they’re easy to spot in git log.

Variation: type errors

Same pattern with mypy:

- id: aider-fix-types
  entry: bash -c 'mypy src/ || aider --no-stream --message "Fix mypy errors. Run mypy src/ to verify."'

When NOT to use this

• On a team where commits are sacred — auto-edits will confuse PR reviewers.
• On a security-sensitive codebase — auto-edits to crypto/auth code without human eyes is a recipe for CVEs.

For those cases, run Aider in CI instead, on a separate branch the human reviews.