Automated Review Loops

Set up Candid to automatically fix issues until your code is clean.

Quick Start

Run the loop with default settings:

/candid-loop

This will:

1. Run candid-review to find issues
2. Automatically apply all fixes
3. Re-run the review
4. Repeat until no issues remain (or max 5 iterations)

Setting Up for CI/CD

GitHub Actions

Add to your workflow:

- name: Run Candid Loop
  run: |
    claude /candid-loop --max-iterations 3

Pre-commit Hook

Add to .pre-commit-config.yaml:

- repo: local
  hooks:
    - id: candid-loop
      name: Candid Loop
      entry: claude /candid-loop --categories critical,major
      language: system
      pass_filenames: false

Managing Ignored Issues

When the loop keeps finding the same issue that you’ve decided is acceptable:

Option 1: Ignore During Review (Interactive Mode)

/candid-loop --mode interactive

When prompted, select “Add to ignore list” for false positives.

Option 2: Pre-configure Ignores

Add to .candid/config.json:

{
  "loop": {
    "ignored": {
      "patterns": ["Unicode handling", "timezone"],
      "categories": ["edge_case"]
    }
  }
}

Option 3: Ignore Specific Issues

Find the issue ID in .candid/last-review.json, then add:

{
  "loop": {
    "ignored": {
      "ids": ["a1b2c3d4e5f6"]
    }
  }
}

Team Consistency

Share ignore patterns across your team by committing .candid/config.json:

{
  "tone": "harsh",
  "loop": {
    "mode": "auto",
    "maxIterations": 5,
    "enforceCategories": ["critical", "major"],
    "ignored": {
      "patterns": ["legacy API compatibility"],
      "categories": []
    }
  }
}

Best Practices

Choose the Right Max Iterations

• 3 iterations: Quick pass, catches most issues
• 5 iterations (default): Good balance
• 10 iterations: Thorough, for complex codebases

Start with Critical Issues Only

For large codebases, start narrow:

/candid-loop --categories critical

Then expand:

/candid-loop --categories critical,major

Use Review-Each Mode for Learning

When working with an unfamiliar codebase:

/candid-loop --mode review-each

This shows you each fix one by one with simple Yes/No prompts, helping you understand what issues exist.

Use Interactive Mode for Full Control

When you need to skip batches or build an ignore list:

/candid-loop --mode interactive

This gives you full control with skip, ignore, and batch options.

Troubleshooting

Loop Never Completes

If the loop keeps finding issues:

1. Check if the same issue keeps reappearing (fix might not be persisting)
2. Add persistent false positives to the ignore list
3. Increase --max-iterations if legitimate issues remain

Loop Exits Too Early

If the loop stops before you expect:

1. Check the summary for “Max iterations reached”
2. Increase --max-iterations
3. Ensure fixes are being applied correctly

Issues Keep Coming Back

If fixed issues reappear in subsequent iterations:

1. The fix might introduce a new issue in a different category
2. Check if your linter/formatter is reverting changes
3. Review the fix logic in the skill output