skill-review

The skill-review tool maintains high-quality standards for AI-agent skill sets by performing automated and manual integrity checks. It addresses the common issue of 'skill rot'—where AI agents provide outdated code, non-existent API imports, or stale library patterns—by executing a rigorous nine-phase audit. The process evaluates codebase consistency against official documentation, validates YAML frontmatter, and ensures dependencies remain current. By automating syntactical checks and providing a structured framework for manual verification, the tool prevents hallucinations caused by obsolete development practices. It classifies issues by severity, distinguishing between critical API mismatches and minor formatting errors. Whether upgrading a skill version or preparing a new submission, this process ensures that bundled resources, README files, and code examples remain accurate, functional, and aligned with current production standards.

When to Use This Skill

• •Auditing an AI skill prior to submission to a public repository
• •Fixing hallucinated API patterns in legacy skill codebases
• •Updating dependencies that have exceeded a 90-day staleness threshold
• •Synchronizing multiple documentation files like README and SKILL.md
• •Refactoring codebases when a library moves to a new major version

How to Invoke This Skill

Example prompts that trigger this skill in Claude Code, Cursor, or Antigravity:

• “Check this skill for outdated API patterns
• “Run the 9-phase audit on the current skill
• “Review this directory for breaking changes
• “Scan the skill for stale dependencies and fix them
• “Validate the schema and imports for this project

Pro Tips

• 💡Integrate the automated review scripts (e.g., `./scripts/review-skill`) into your CI/CD pipeline to continuously monitor skill health and catch issues early.
• 💡Prioritize fixes based on the 'Severity' categorization (Critical, High, Medium, Low), focusing first on security vulnerabilities and breaking changes that impact core functionality.
• 💡Leverage the 'Auto-fix unambiguous' feature for efficient resolution of minor issues, but always perform a human review for architectural or design-related changes.

What this skill does

• •Execute automated 9-phase audits for code and configuration accuracy
• •Identify and resolve non-existent API imports and method signature drift
• •Validate YAML frontmatter and directory structure for discovery optimization
• •Generate standardized audit reports with severity-based classification
• •Perform version control management including semantic versioning bumps

When not to use it

• ✕Running on non-technical documents that lack API or schema definitions
• ✕Performing initial conceptual brainstorming where structure is not yet defined

Example workflow

1. Invoke /review-skill <skill-name> to initiate the process
2. Allow the automated script to validate YAML and link integrity
3. Review the audit report to identify critical API mismatches
4. Apply auto-fixes for unambiguous issues like missing imports
5. Manually verify architectural choices for high-severity findings
6. Commit the updated files with a generated changelog

Prerequisites

• –Active AI-agent development environment
• –Access to the target skill repository
• –Installed review-skill bash scripts

Pitfalls & limitations

• !Auto-fixes may not resolve complex architectural inconsistencies
• !Requires manual intervention for breaking changes that impact logic
• !False positives can occur if local documentation is not up to date

FAQ

How it compares

Unlike a generic code review prompt, this tool utilizes a rigid 9-phase protocol specifically calibrated for AI agent metadata, bundled resource integrity, and external documentation syncing.

Source & trust

Recommended Rules

Recommended Workflows

Recommended MCP Servers