How does the tool determine a PASS or FAIL verdict?

Verdict logic is tiered: any Critical failure results in a FAIL; 3 or more Important failures result in NEEDS WORK (or FAIL if --strict is used); fewer failures result in a PASS.

Can it automatically fix my SKILL.md file?

Yes, in its default mode, the tool identifies failures and uses a multi-select prompt to ask which issues you would like it to automatically fix for you.

What kind of security checks does it perform?

It scans for shell-injection vulnerabilities (such as shell=True), usage of unsafe functions like eval, exposure of secrets or credentials, and improper file permissions.

Can I run an audit without making changes to my files?

Yes, by using the --dry-run flag, you can generate a full audit report and verdict without the tool performing any Edit or Write operations.

What is the primary purpose of the Review Skill?

The Review Skill is designed to audit and fix Claude Code skill definitions (SKILL.md) to ensure they follow security, structural, and performance best practices defined by Anthropic and the community.

Claude Skill Reviewer

Name: Claude Skill Reviewer
Author: smykla-skalski

bysmykla-skalski

•

安全与测试

Audits and automatically fixes Claude Code skill definitions based on Anthropic best practices and security guidelines.

The Review Skill is a comprehensive quality assurance tool designed to ensure Claude Code skill definitions (SKILL.md) meet rigorous standards. It evaluates skills against a tiered checklist covering critical security, performance, and structural requirements. By automating the validation of frontmatter, file references, security vulnerabilities, and script configurations, it provides a clear PASS/FAIL verdict. It doesn't just find problems; it can also intelligently apply fixes to bring your skills into compliance with the Agent Skills specification and community standards.

主要功能

01Intelligent 'Fix Mode' that applies corrections to SKILL.md and bundled resources with user approval

02Tiered binary checklist (Critical, Important, Polish) for structured quality assessment

03Automated security scanning for shell-injection risks, secret leaks, and unsafe code patterns

042 GitHub stars

05Flexible reporting options including Markdown and JSON for integration into development workflows

06Validation of script permissions, exit codes, and cross-file dependency declarations

使用场景

01Standardizing an existing internal skill library to match Anthropic's recommended patterns

02Auditing a new Claude Code skill before publishing to ensure it meets community and security standards

03Automating security and linting checks for skill definitions within a CI/CD pipeline

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add smykla-skalski/sai review-skill

For use in Claude.ai and ChatGPT