tracelens-analysis-orchestrator
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads setup and configuration scripts from well-known technology services. Evidence in evals/eval_scripts/run_pi_analysis_in_docker.sh and evals/README.md shows fetching Node.js setup scripts from NodeSource (deb.nodesource.com, rpm.nodesource.com) and the Cursor CLI installation script from cursor.com.
- [REMOTE_CODE_EXECUTION]: The evaluation harness contains patterns for downloading and executing scripts via pipe-to-bash for environment setup. For example, run_pi_analysis_in_docker.sh installs Node.js using official NodeSource distribution scripts. These operations target well-known services and are consistent with the setup requirements of the evaluation harness.
- [COMMAND_EXECUTION]: The orchestrator is designed to execute a sequence of analysis scripts across local and remote environments. It builds command prefixes using user-supplied parameters (node, container, venv) to run analysis tools via SSH and Docker exec. This behavior is the primary intended function of the skill for performance orchestration.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external PyTorch profiler traces, which represents a potential vulnerability surface. Ingestion points occur in orchestrator_prepare.py where trace files enter the context. The skill uses structured templates and specific HTML-style markers () to delimit output, and trace data is parsed into structured metrics before agent processing to mitigate potential injection risks.
- [DYNAMIC_EXECUTION]: The evaluation scripts use dynamic command execution for service management. The run_pi_analysis_in_docker.sh script uses the eval command to start an inference server using a user-provided command string within a controlled Docker environment.
Audit Metadata