The Agent Skills Directory

Command Execution: The skill initiates security pipelines by executing local Python and shell scripts (launch.py, run.sh) located within the defined Mantis installation directory. This is the intended mechanism for starting campaigns.\n- Environment Auto-Configuration: The tool can automatically detect environment settings, such as Google Cloud project IDs using the gcloud CLI or virtualization capabilities through /dev/kvm. These operations are used to ensure the review pipeline has the necessary configuration to run effectively.\n- Indirect Prompt Injection Surface: As the skill is designed to scan untrusted code for vulnerabilities, it possesses an inherent indirect prompt injection surface.\n
Ingestion points: The skill reads target source code files or entire repositories provided during the launch command.\n
Boundary markers: The skill instructions do not specify explicit delimiters; boundary management is typically handled by the underlying analysis agents.\n
Capability inventory: The pipeline can execute scripts and perform network requests to AI model providers (Vertex AI, OpenAI-compatible endpoints).\n
Sanitization: Sanitization of the scanned content is delegated to the security engine's analysis logic.\n- Network Interaction: The skill supports connecting to external AI model endpoints through flags like --api-base, which involves sending data to user-specified services for analysis.

mantis-launch