rl-env-from-description
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided natural language descriptions to generate and scaffold executable RL environment code. This ingestion of untrusted data into code-generation templates poses a risk where malicious input could lead to unsafe code generation.\n
- Ingestion points: Environment descriptions in SKILL.md.\n
- Boundary markers: None identified in the prompt templates to separate user data from instructions.\n
- Capability inventory: Includes file-writing for multiple frameworks and the execution of generated rollout scripts (SKILL.md Step 4).\n
- Sanitization: No input validation or sanitization of the description is described.\n- [DYNAMIC_EXECUTION]: The skill uses Step 4 (Validation) to run a generated
rollout.pyscript against a small LLM. This involves executing code created at runtime from user descriptions.\n- [COMMAND_EXECUTION]: The validation flow executes shell commands to run the smoke tests for each framework variant.\n- [EXTERNAL_DOWNLOADS]: The documentation provides instructions to install thenemo_gymlibrary from NVIDIA's official GitHub repository.
Audit Metadata