The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the shell tool to execute bash loops for file enumeration, token counting via wc, and gbrain CLI commands for system configuration retrieval and cross-modal evaluation.
[DATA_EXFILTRATION]: The skill reads several sensitive local files including ACCESS_POLICY.md and USER.md. While the stated purpose is token-hygiene auditing, this behavior exposes identity details and access rules to the agent's context.
[PROMPT_INJECTION]: The skill processes untrusted content from files like MEMORY.md and CLAUDE.md. This ingestion surface is vulnerable to indirect prompt injection, where malicious instructions embedded in these files could attempt to influence the audit findings or the behavior of the evaluation model during the judging phase.

context-audit