The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill compiles and executes code from the target repository during the fuzzing phase (Arm A3.5) and the PoC verification gate (Arm A4.5). Although it attempts to mitigate risk by scrubbing sensitive credentials from the environment using 'env -u', the execution of code from untrusted third-party sources represents a significant security risk.
[REMOTE_CODE_EXECUTION]: The skill fetches the 'osv-scanner' binary from a known organization's GitHub releases at runtime. While this is performed for legitimate auditing purposes and uses a well-known source, it involves executing externally retrieved binaries.
[COMMAND_EXECUTION]: The skill installs and invokes multiple security auditing tools (e.g., semgrep, slither-analyzer, trufflehog) via the shell. It specifically manages the PATH and uses bare-name invocations to satisfy platform execution constraints.
[DATA_EXFILTRATION]: The skill is designed to send vulnerability reports to external destinations. It uses the GitHub API for Private Vulnerability Reporting and the Resend API for outbound email disclosures. It also connects to external DNS-over-HTTPS services to verify recipient email domains.
[INDIRECT_PROMPT_INJECTION]: By auditing untrusted source code and metadata, the skill is exposed to potential malicious instructions intended to subvert the auditing logic. The skill incorporates a mandatory reproduction gate for high-severity findings to minimize the impact of such injections.

vuln-scanner