The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local shell script (bin/skill-scan.sh) to gather data about the filesystem and installed agent components.
[DATA_EXPOSURE]: The scanning script accesses multiple agent-specific configuration and skill directories in the user's home folder (including .claude, .cursor, .codex, and .agents). It reads SKILL.md and openai.yaml files to extract metadata, versions, and usage patterns. This data is used solely for local auditing and is not exfiltrated.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process data from other, potentially untrusted, agent skills. This creates an attack surface where a scanned skill could contain malicious instructions designed to influence the auditor's judgment.
Ingestion points: Reads content from SKILL.md and openai.yaml files in all detected skill directories.
Boundary markers: Absent; the script parses and passes raw metadata/content previews to the agent.
Capability inventory: Shell access for the scanner script; the agent performs technical analysis of the output.
Sanitization: No explicit prompt-level escaping is described for the interpolated scan results.

skill-hygiene