The Agent Skills Directory

Command Execution: The skill initiates its analysis pipeline by executing local scripts, including ./reference/run.sh and python3 reference/scripts/launch.py. This is the intended functional mechanism for the tool's operation.\n- Indirect Prompt Injection: Because the skill reads and analyzes external source code files, it is subject to the risk of indirect prompt injection, where malicious instructions within the scanned code could attempt to influence the AI model's behavior. This is a common consideration for automated code analysis tools.\n
Ingestion points: Target source code files and directory contents specified by the user.\n
Boundary markers: The instructions do not describe specific delimiters or safety warnings for the ingested content.\n
Capability inventory: Execution of local scripts, database writes (knowledge.db), and network communication with LLM endpoints.\n
Sanitization: No specific filtering or sanitization steps for the scanned code are detailed in the instructions.\n- Environment Access: The skill includes an 'Auto-Healing' feature that detects host environment capabilities, such as GCP project IDs via gcloud and virtualization support via /dev/kvm, to facilitate automated configuration.

mantis-launch