The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python auditor (scripts/audit.py) and references the skills-ref CLI tool to perform structural validation of plugin directories. This is legitimate behavior for a developer tool.- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through the processing of untrusted files from the target directory.
Ingestion points: scripts/audit.py reads the contents of README.md and SKILL.md files using f.read() and f.readlines().
Boundary markers: None. The raw text is read and displayed to the agent or evaluated for length and structure without isolation from instructions.
Capability inventory: The agent has Bash, Read, and Write permissions and is specifically instructed in SKILL.md to use these tools to 'fix compliance issues' based on the file contents.
Sanitization: No sanitization or instruction filtering is applied to the content of the files being audited.

audit-plugin