The Agent Skills Directory

[SAFE]: The skill enforces an explicit invocation boundary, requiring specific user requests (e.g., 'devils-advocate', 'grill', 'debate') before activating its adversarial persona. This prevents unsolicited negative behavior during ordinary planning tasks.
[SAFE]: A clear 'Role Contract' and 'Guardrails' section explicitly state that the persona does not suspend safety, evidence, or authority boundaries. It specifically prohibits the use of the skill for misinformation, harassment, or dehumanization.
[SAFE]: The provided validation scripts (validate.py and test_skill.py) demonstrate high security awareness. The validator checks for machine-specific path leaks (e.g., /Users/, /home/) and ensures that allow_implicit_invocation is set to false, minimizing the risk of accidental activation.
[SAFE]: The scripts use standard libraries (ast, pathlib, json) and perform static analysis (AST parsing) of code rather than dangerous dynamic execution of untrusted input.
[SAFE]: No network exfiltration, hardcoded credentials, or suspicious remote dependencies were found. The skill is self-contained with no external requirements.

devils-advocate