The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform network diagnostics and security posture checks. Access is restricted to specific passive wrappers (such as curl -sI for headers or openssl s_client) via internal instructions and an intended settings.json contract that explicitly denies raw offensive tools like sqlmap, nmap (raw), or hydra.
[DATA_EXFILTRATION]: The skill incorporates robust data handling policies that mandate the redaction of PII, session tokens, and credentials from all outputs. It specifically prohibits the capture of raw request/response bodies when performing diagnostic probes to prevent accidental exposure of sensitive data.
[INDIRECT_PROMPT_INJECTION]: The skill includes procedures for validating AI agent boundaries against prompt injection. These activities are restricted to 'observation-only' modes using benign canary tokens, with strict termination conditions and a prohibition on utilizing injection triggers for real unauthorized actions.
[SAFE]: Static detection of a destructive command (rm -rf /) is a confirmed false positive. The command appears exclusively within a list of explicitly forbidden 'Restricted Activities' in Section 4 of the documentation, serving as a safety constraint rather than an executable instruction.

authorized-pentest-validation