The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and evaluate project-specific rule files.
Ingestion points: Scans and reads content from .md files in the .claude/rules/ directory.
Boundary markers: No delimiters or boundary instructions are mentioned to prevent the agent from obeying instructions found inside the rule files.
Capability inventory: The skill analyzes file content for quality metrics and frontmatter validity.
Sanitization: No sanitization of rule content is described in the provided validation modules.
[SAFE]: No other security issues such as hardcoded credentials, obfuscation, or unauthorized network activity were detected in the documentation and configuration files provided. The skill's primary logic script (scripts/rules_validator.py) was not included in the analyzed fileset, and evaluation was performed on the provided instructional modules.

rules-eval