The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on executing local shell scripts (./python.sh, tracy-capture) and complex command lines. In the unit testing workflow, the agent is instructed to extract a command string from the stdout of a previous test run and execute it directly. This pattern of capturing and re-executing data from logs carries a risk if the log content can be influenced by external factors.
[EXTERNAL_DOWNLOADS]: The skill configuration references the download of simulation assets from an Amazon S3 bucket (https://omniverse-content-staging.s3-us-west-2.amazonaws.com/Assets/Isaac/6.0). This is a well-known service used for legitimate asset distribution.
[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes data from externally generated CSV and log files, which represents an indirect injection surface.
Ingestion points: The script scripts/compare_tracy_csvs.py reads data from CSV files specified by the user or benchmark output; metrics are also extracted from .log files generated by benchmarks.
Boundary markers: No explicit delimiters or warnings to ignore instructions within the processed data are present in the scripts or instructions.
Capability inventory: The skill possesses capabilities to execute shell commands, read and write to the local file system (e.g., /tmp/tracy_profiles), and perform network operations for asset loading.
Sanitization: The comparison script parses CSV fields such as 'name' and prints them to stdout for the agent to read without performing sanitization or escaping of the content.

profile-isaac-sim