The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill clones the Isaac for Healthcare workflows repository from https://github.com/isaac-for-healthcare/i4h-workflows if the repository is not already present in the expected local path. This is a standard setup step for accessing the necessary training scripts and environment configurations.
[COMMAND_EXECUTION]: The instructions require the agent to execute several local shell scripts, including ./train.sh and ./run.sh, to manage the training, evaluation, and export lifecycle of RL policies. These commands are integral to the skill's primary function of training robotic models.
[INDIRECT_PROMPT_INJECTION]: The skill processes declarative YAML configurations and Scene manifests which could be manipulated by an attacker if the repository content is compromised.
Ingestion points: The agent is instructed to read profiles under ./rl/profiles/, trainer configurations under ./rl/config/, and Scene manifest implementations.
Boundary markers: The instructions do not specify the use of delimiters or specific security warnings to the agent regarding instructions that might be embedded within these configuration files.
Capability inventory: The skill has the capability to read files from the local system and execute commands via the ./train.sh and ./run.sh entry points.
Sanitization: The process includes a --dry-run phase intended to verify source existence and basic schema consistency, but it does not detail robust sanitization or sandboxing of the interpreted configuration data.

i4h-workflow-train-rl