The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions focus exclusively on security auditing. No patterns attempting to bypass safety filters, extract system prompts, or override core agent behavior were detected.
[DATA_EXFILTRATION]: No network operations or data exfiltration patterns are present. References to sensitive file paths (e.g., /etc/passwd, .ssh/id_rsa) and secrets in the documentation and examples are used as test cases for the skill's detection capabilities, not for unauthorized access.
[CREDENTIALS_UNSAFE]: No hardcoded credentials or secrets belonging to the skill or its author were found. Example secrets in testing scenarios (e.g., 'hunter2-prod') are clearly marked as placeholders or examples for detection.
[REMOTE_CODE_EXECUTION]: The skill does not execute code dynamically or fetch remote scripts. It provides a list of recommended security tools (e.g., Semgrep, Bandit, CodeQL) for users to integrate into their own development environments.
[COMMAND_EXECUTION]: No unauthorized command execution or privilege escalation patterns were found. The skill instructions analyze command injection vulnerabilities in target code but do not invoke shell commands themselves.
[EXTERNAL_DOWNLOADS]: The skill references documentation and research from trusted organizations including OWASP, NIST, and MITRE. No untrusted binary or script downloads are initiated.

sweeping-for-security