The Agent Skills Directory

[SAFE]: The skill is a utility for generating DeepStream GStreamer pipelines. It follows security best practices and contains no malicious code or exfiltration patterns.
[COMMAND_EXECUTION]: The validate_pipeline.py script and the test suite utilize subprocess.run to interact with the local NVIDIA DeepStream SDK installation (specifically gst-inspect-1.0 and gst-launch-1.0). This is used to verify element existence and perform dry-run syntax checks.
The implementation uses list-form arguments (shell=False) to prevent shell injection.
Element names extracted from user queries are validated against a strict alphanumeric regex ([a-zA-Z0-9_-]) before being passed to subprocess calls.
[EXTERNAL_DOWNLOADS]: The skill is designed to operate with zero external dependencies, using only the Python standard library. No remote package installations or script downloads were detected.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill performs local file system checks (via ls) to verify the presence of DeepStream sample streams and configuration files (e.g., /opt/nvidia/deepstream/...). These checks are limited to the SDK environment and do not access sensitive user directories or credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes user natural language queries to construct search terms for its BM25 retrieval engine. The risk of malicious user input affecting the agent is mitigated by a structured requirement-extraction workflow using a fixed question bank and mandatory output formatting rules that prevent the agent from executing arbitrary scripts created by the user input.

deepstream-generate-pipeline