The Agent Skills Directory

[SAFE]: The skill documents security vulnerabilities related to bypassing human approval gates for educational and red-teaming purposes. It describes techniques such as summary manipulation and approval fatigue to help security professionals test agent robustness.
[COMMAND_EXECUTION]: Includes a self-contained Python script designed for auditing. The script uses standard libraries (json, urllib.request) to compare an agent's reported summary with its actual tool execution. It interacts only with user-defined target placeholders and contains no malicious payloads or exfiltration logic.

hunt-agent-hitl-bypass