The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to reference and link to official NVIDIA Isaac Sim documentation and examples located at docs.isaacsim.omniverse.nvidia.com and within the local $ISAAC_SIM_DIR.
[COMMAND_EXECUTION]: Provides instructions to use the grep command locally to search the skill library (skills/) for existing concepts to avoid duplicate entries during the distillation process.
[INDIRECT_PROMPT_INJECTION]: The distillation process creates a feedback loop where user corrections and interaction history are ingested to update the agent's future instructions.
Ingestion points: The agent reviews memory/ files and user feedback history.
Boundary markers: Not explicitly defined, but the skill enforces a 'Generalization Rule' to transform specific user input into broader, safer procedural guidelines.
Capability inventory: The skill possesses the capability to write and modify files within the skills/ and scripts/ directories.
Sanitization: Relies on the agent's internal reasoning and verification steps to ensure new instructions are procedural and generalized rather than verbatim reflections of untrusted input.
[DATA_EXFILTRATION]: No patterns of unauthorized data exfiltration or access to sensitive system credentials were detected. The skill's file operations are scoped to its own library and workspace scripts.

skill-distillation