moos-ivp-eval-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 utilizes shell scripts (launch.sh, zlaunch.sh, moos_scoped_teardown.sh) to orchestrate MOOS-IvP simulations. These scripts execute domain-specific tools like pAntler and nsplug, as well as system utilities like kill and lsof. The use of these commands is strictly aligned with the skill's purpose of mission automation and process management.
  • [SAFE]: The moos_scoped_teardown.sh script demonstrates a security-conscious design by ensuring that process termination is scoped to the mission's directory. By checking the working directory of target processes, it prevents the disruption of unrelated system activity.
  • [EXTERNAL_DOWNLOADS]: The skill documentation and scripts reference an external dependency, xlaunch.sh. While not provided within the skill files, this utility is a standard component of the MOOS-IvP evaluation infrastructure and is expected to be pre-installed in the target environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill includes scripts for static analysis of mission configuration files. These scripts use grep, sed, and awk to validate parameters. This ingestion of external data is limited to structural validation and does not pose a risk of instruction override or injection into the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 11:28 PM