trt-perf-analysis

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses subprocess.run() in scripts/package_report.py to orchestrate internal scripts for data analysis and validation. These executions are limited to internal skill files and use the current Python interpreter.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted TensorRT JSON profiles.
  • Ingestion points: User-provided JSON files are loaded in scripts/analyze_trt_perf.py.
  • Boundary markers: Data is structured JSON; no explicit natural language delimiters are used.
  • Capability inventory: The skill can execute internal scripts (scripts/package_report.py) and write files to the filesystem.
  • Sanitization: Inputs are validated against a formal JSON schema in scripts/validate_analyze_data.py. The frontend report uses DOMPurify in report_template/app.js to sanitize Markdown content before rendering.
  • [DYNAMIC_EXECUTION]: The script scripts/trt_perf/layer_info.py uses importlib.import_module() to load visualization targets. This is restricted to a hardcoded list of internal module paths, posing no risk of loading untrusted code.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 11:34 AM
Security Audit — agent-trust-hub — trt-perf-analysis