ron-myers/candid

Overview

This skill runs candid-review repeatedly until your code meets the configured quality gates. It automates the fix → review → re-run cycle with three modes (auto, review-each, interactive) and supports ignored categories, patterns, and issue IDs. Use CLI flags or config files to control behavior and limits.

How this skill works

On start it loads loop settings from CLI flags, project config (.candid/config.json), or user config (~/.candid/config.json), falling back to sensible defaults. It then runs a loop up to maxIterations: invoke candid-review, read the saved review state, filter issues by enforced categories and ignores, and apply or skip fixes according to the selected mode. The loop stops when no enforceable issues remain, when the user cancels, or when the iteration limit is reached.

When to use it

• Automatically clean up routine lint, style, and simple security fixes across a codebase.
• Iteratively apply changes when many small fixes are needed but you want control over specific changes.
• Enforce only selected categories (e.g., critical and major) while skipping known false positives.
• Start a safe, repeatable cleanup during CI or before merging large branches.
• Maintain an interactive review workflow for complex or debated fixes.

Best practices

• Start with auto mode for fast bulk cleanup, then switch to review-each or interactive for fine-grained control.
• Use enforceCategories to limit the loop to high-impact issues first (critical, major).
• Add persistent ignores (categories, title patterns, or IDs) for known false positives.
• Keep maxIterations reasonable (5–10) to avoid long or infinite loops.
• Review the final summary and remaining issues to decide on manual action or config changes.

Example use cases

• Run /candid-loop after a large refactor to automatically apply small fixes across files.
• Use review-each mode when onboarding to learn what candid-review changes before applying them.
• Run interactive mode to add recurring false positives to the ignore list while fixing the rest.
• Limit to critical categories in CI to block dangerous regressions while leaving stylistic fixes for later.
• Combine --max-iterations and --categories to run focused, bounded cleanup passes.