moos-ivp-mission-builder
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes shell scripts (launch.sh, launch_vehicle.sh, launch_shoreside.sh) and validation scripts (check_generated_ports.sh, check_generated_networking.sh) that execute MOOS-IvP tools such as nsplug, pAntler, and uMAC. This execution is essential for building and validating robotics missions. The scripts implement proper process group termination using 'kill -- -$$' to ensure clean shutdown of all background components.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests local mission configuration files (.moos, .bhv) to generate deployment targets. 1. Ingestion points: Reads mission files and directory structure provided in the MISSION_DIR argument across several files like scripts/static_check_mission.sh. 2. Boundary markers: None; the skill relies on the structural integrity of the MOOS-IvP configuration language. 3. Capability inventory: Includes target file generation, shell command execution for mission launching, and file system cleanup. 4. Sanitization: Standard shell variable expansion is used without specific sanitization for content embedded within configuration files. This represents a low surface for indirect injection given the specific technical use case.
- [SAFE]: The skill demonstrates proactive quality and security checks via static_check_mission.sh, which identifies potentially disruptive global process kills (e.g., ktm or pkill) and encourages portable scripting. No credential exposure, obfuscation, or unauthorized network operations were detected.
Audit Metadata