The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive global and project-specific configuration files, including ~/.claude/settings.json and the content of all installed skill files in ~/.claude/skills/. This data access is essential for calculating the total token overhead of the agent's environment, and the skill does not perform any network operations to send this data externally.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading and analyzing instruction files (CLAUDE.md, RULE.md, SKILL.md). Although the skill only audits the text for length and specific patterns rather than executing it as commands, this ingestion constitutes an indirect prompt injection surface.
Ingestion points: Reads project-level and global instruction files (CLAUDE.md, RULE.md, SKILL.md) to apply bloat filters.
Boundary markers: The skill does not implement specific boundary markers or "ignore embedded instructions" warnings when processing the content of these files.
Capability inventory: The skill has the capability to modify the local settings.json file and provides the agent with the ability to rewrite instruction files based on its audit findings.
Sanitization: There is no evidence of content sanitization, escaping, or validation performed on the ingested text before analysis.

usage-audit