skill-scanner

To audit SKILL.md files for prompt injection and scope drift, run a scanner against the target files to detect injection phrases, frontmatter validity, and body length inconsistencies. It returns structured findings with severity and actionable fixes for review.

Description drift in agent context occurs when SKILL.md body content diverges from its frontmatter scope. It is detected through consistency checks that validate frontmatter fields against the body text, preventing unscoped authority and misaligned agent instructions.

Yes, you can use a SKILL.md scanner as a CI quality gate. It generates a CI-ready findings report containing severity levels, evidence, and actionable fixes to guide skill merging decisions and enforce safe agent context during integration.

Yaml frontmatter validation in skill documentation is performed by scanning the SKILL.md file structure. The scanner checks for missing sections, verifies frontmatter field consistency against the body content, and reports findings on structural validity to ensure proper documentation.

Automated prompt-injection detection in skill files relies on identifying known injection phrases and unscoped authority patterns. Limitations include potential false positives from legitimate instructional text and the inability to detect novel or highly obfuscated injection vectors outside predefined detection rules.