sim2real-iteration
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute
python -m robosim_gauntlet.clifor preflight audits of physics models and comparing different controller candidates. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: The analysis loop processes untrusted data from simulation logs and metrics, specifically
result.grade.violations,result.grade.metrics, andresult.log(SKILL.md). - Boundary markers: There are no explicit delimiters used to separate the ingested simulation data from the agent's instructional context.
- Capability inventory: The skill possesses the capability to execute shell commands (
robosim_gauntlet.cli) and launch controller candidates (episode.fork). - Sanitization: No content filtering or sanitization is mentioned for the processed logs, though the data is primarily numerical or structured metrics.
- [DATA_EXFILTRATION]: The skill uses
os.environ["GAUNTLET_URL"]to establish communication with a remote device layer viaproxy.RemoteDeviceLayer. This network activity is a functional requirement for interacting with remote robotics controllers.
Audit Metadata