metric-validation-harness

Pass

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The harness executes a user-provided command string (via the METRIC_CMD configuration or environment variable) to evaluate metric performance on various code samples. This is the intended purpose and documented behavior of the tool.
  • [COMMAND_EXECUTION]: The skill uses shell execution in Bash scripts (verify.sh, check-determinism.sh, etc.) and subprocess.run in Python scripts (harness.py, check-tractability.py) to orchestrate the validation experiments. Python calls use shlex.split to mitigate risks associated with command-line argument construction.
  • [SAFE]: The skill is read-only and explicitly avoids modifying the metric code, the corpus artifacts, or any system state. No network activity or external downloads are performed.
  • [SAFE]: No obfuscation, persistence mechanisms, hardcoded credentials, or prompt injection patterns were detected. The harness implementation is transparent and relies exclusively on standard library modules.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 6, 2026, 09:03 PM
Security Audit — agent-trust-hub — metric-validation-harness