The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process user-provided Isaac Gym source code, including task definitions, reward functions, and training configurations. This untrusted content is then executed by the agent to perform validation and training tasks.
Ingestion points: Untrusted Isaac Gym environment code and configuration files provided by the user for migration (e.g., IsaacGymEnvs tasks and YAML configs referenced in reference.md).
Boundary markers: The instructions do not define delimiters or specific warnings to prevent the agent from following instructions potentially embedded within the user-provided code.
Capability inventory: The skill utilizes uv run python, pytest, and training runners (isaaclab train) to execute the migrated code as part of its validation loop in SKILL.md and reference.md.
Sanitization: There are no requirements for the agent to sanitize or perform a safety review of the user-provided scripts before execution.
[COMMAND_EXECUTION]: The skill frequently instructs the agent to execute shell commands to validate the environment, run tests, and perform policy evaluations.
Evidence: SKILL.md includes a runtime preflight check: uv run --extra isaacsim python -c "import importlib.util, sys; ..." and testing commands like uv run --with pytest python -m pytest PATH_TO_MIGRATION_TEST.
Evidence: reference.md describes a policy success validation loop that involves multiple command-line operations, including uv run ... isaaclab train and checkpoint evaluation scripts.
Context: These commands are used to execute logic that is directly derived from the user's migration target, increasing the impact of the indirect prompt injection surface.

isaaclab-migrating-from-isaac-gym