Skill Inspector Tier Two

Overview

This agent acts as a senior qualitative reviewer for AI skills. It is designed to take skills that have already passed initial, mechanical validation (Tier 1) and subject them to a rigorous human-level assessment. The goal is to ensure the skill is not only functional but also high-quality, safe, and maximally usable.

Capabilities

• Content Quality Scoring: Rates the skill's documentation (SKILL.md) on a 1-5 scale based on focus, completeness, and clarity of purpose.
• Actionable Trigger Detection: Awards points if the description contains clear, actionable trigger phrases that tell an agent when to use the skill.
• Instruction Clarity Check: Assesses the instructions for ambiguity or contradictions, awarding partial credit for unambiguous logic.
• Prompt Injection Security Audit: Scans the entire skill content for known prompt injection patterns, including attempts to override system prompts or exfiltrate data.
• Structured Merging: Consolidates all findings (scores and notes) into existing Tier 1 NDJSON records for comprehensive auditing.

Example Use Cases

• Pre-Release Vetting: Before a skill is published, use this agent to ensure its documentation meets the highest standards of usability and safety.
• Competitor Analysis: Reviewing skills from other sources to benchmark best practices in prompt engineering and documentation structure.
• System Hardening: Periodically running this against an existing catalog of internal skills to proactively identify subtle security vulnerabilities or outdated instructions.