ridewind/my-skills
Overview
This skill orchestrates a full code review execution pipeline. It loads and validates review configuration, collects code context (diffs, commits, branches), starts multiple parallel subagents using a chosen preset, and produces a consolidated report. It supports branch vs branch, MR/PR, and full-project reviews with optional debug tracing.
How this skill works
The executor finds and merges configuration files (project, user, global), validates presets, and prompts the user to pick a review preset. It collects code artifacts (diff.patch, commits.json, branch-info.json) into a timestamped working directory, asks for confirmation, then launches configured subagents in parallel. When subagents finish, it aggregates, deduplicates, and classifies findings into a comprehensive summary and per-skill reports.
When to use it
- When you ask “review my code” or “do a code review”
- When you need a branch review or branch A vs branch B comparison
- When you request a PR/MR review by number or URL
- When you want a preset-based, multi-skill review (security, performance, style)
- When you need an auditable report with optional debug logs
Best practices
- Keep project-level config up to date so presets reflect available skills
- Confirm the collected code context before launching subagents
- Use named presets (quick, full, security-first) to control scope and runtime
- Enable debug mode only when troubleshooting to capture full decision logs
- Review per-skill reports in addition to the consolidated summary for traceability
Example use cases
- Review a feature branch before merging (feature/auth vs dev) and get a prioritized issue list
- Run a security-first preset on a PR to surface potential injection or credential leaks
- Run a full preset on a large refactor to combine style, performance, and security checks
- Audit an entire monorepo by selecting specific project paths and receiving per-project reports
- Re-run a failed review with debug mode enabled to capture subagent failures and timings
FAQ
A timestamped working directory with code-context.json, diff.patch, commits.json, branch-info.json, DEBUG-SESSION.md (if debug enabled), reports/, and the comprehensive summary file.
How are conflicts between multiple skill findings handled?
Findings are deduplicated by file path, line/range, and type; duplicates are merged and the originating skills are recorded. Issues are then bucketed by severity (Critical, High, Medium, Low, Info).
What happens if a subagent fails?
Failures are logged to the debug session and the executor continues waiting for other subagents. The final summary marks any skill that failed and includes failure details.
