The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Tracy CSV exports and test runner logs, which are subsequently incorporated into the agent's execution context. This creates a potential surface where malicious strings embedded in profiling data (e.g., zone names) could influence the agent.
Ingestion points: The scripts/compare_tracy_csvs.py script reads and parses user-supplied CSV files; the 'Profiling Extension Unit Tests' workflow in SKILL.md requires the agent to parse the stdout of a test runner to extract command lines.
Boundary markers: No explicit delimiters or instructions to ignore embedded content are used when presenting the parsed data or extracted strings to the agent.
Capability inventory: The skill leverages the run_script helper and instructs the agent to execute various shell commands (e.g., ./python.sh, tracy-capture) within the simulator environment.
Sanitization: While scripts/compare_tracy_csvs.py performs type validation on numeric metrics, the profiling zone names are handled as raw strings and printed in the comparison report without sanitization.
[EXTERNAL_DOWNLOADS]: The skill references an S3 bucket for downloading assets (https://omniverse-content-staging.s3-us-west-2.amazonaws.com/Assets/Isaac/6.0). This is an official resource for the simulator hosted on a well-known cloud service provider.
[COMMAND_EXECUTION]: The skill defines procedures for executing shell scripts and binaries, such as ./python.sh and tracy-capture, to facilitate performance measurement. These operations are essential to the skill's primary function and are limited to the simulator's directory.

profile-isaac-sim