The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains strings like "ignore previous instructions" and "disregard your system prompt." These are provided as examples of malicious patterns for the agent to detect in external files, not as instructions to override its own behavior. This is a false positive for prompt injection.
[COMMAND_EXECUTION]: The skill uses tools like Read, Grep, and Glob to inspect file contents. It does not execute the files or commands it finds; its purpose is purely analytical.
[DATA_EXFILTRATION]: The skill lists sensitive file paths (e.g., ~/.ssh, .env) as indicators of potential risk in the files being audited. It does not attempt to access these files on its own behalf.
[SAFE]: The configuration and instructions are consistent with the skill's stated purpose of being a security scanner. Access to file reading tools is necessary and restricted by the provided frontmatter.

agent-guard