The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/batch_actor_sdg.py uses subprocess.run to execute simulation tasks. Security is maintained by passing the command as a list of strings and avoiding shell interpretation, which prevents typical command injection attacks. The paths for the launcher and scripts are provided via command-line arguments.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data which could theoretically be used to influence agent behavior.
Ingestion points: The generator script reads base.yaml and sweep.yaml, and the batch runner reads manifest.csv to identify configuration variants.
Boundary markers: The instructions do not define specific delimiters or "ignore previous instructions" markers for the content within the processed YAML or CSV files.
Capability inventory: The skill possesses capabilities for file system modification (creating and deleting variant directories matching specific patterns) and subprocess execution of local scripts.
Sanitization: Input YAML files are parsed using yaml.safe_load. The skill also implements optional JSON Schema validation using the jsonschema library to ensure that generated configurations adhere to the expected Isaac Sim Replicator Agent structure.

actor-sdg-sweep-config