isaaclab-planning-manipulation-tasks
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on how to interpret and respond to user queries regarding robotics task planning (e.g., in
evaluations.md). This creates a standard surface for indirect prompt injection where a user might embed instructions within a task description. However, the skill's emphasis on physical validation, phase-based decomposition, and deterministic metric checks provides a robust framework that prioritizes simulation integrity. - Ingestion points: User queries provided in
evaluations.md(e.g., Scenario 1: "Help me create a grasp-and-place task..."). - Boundary markers: None explicitly defined.
- Capability inventory: The skill defines logic for planning and validation; it does not grant the agent capabilities for network operations, arbitrary file writes, or subprocess execution based on user input.
- Sanitization: Not specified.
- [COMMAND_EXECUTION]: The skill documentation includes a command for maintainers to validate changes (
uv run --no-project python tools/skills/cli.py check). This command relies on a local environment script (tools/skills/cli.py) and follows standard developer practices for skill maintenance without fetching or executing untrusted remote content.
Audit Metadata