tao-run-deft-pas

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: Orchestrates model training, evaluation, and data mining tasks via docker run and subprocess.run. Execution is strictly governed by an immutable transaction log and requires all commands to match pre-approved digests.
  • [DYNAMIC_EXECUTION]: Implements a secure execution shim in run_deft_cli.py using os.execve to run specific toolkit binaries within a validated environment. It also utilizes runtime import hooks in patches/sitecustomize.py to ensure internal compatibility between data types.
  • [INDIRECT_PROMPT_INJECTION]: Processes external natural language datasets containing image captions. Risks are mitigated through rigorous schema validation of all inputs and the isolation of the training workload from the control plane.
  • [EXTERNAL_DOWNLOADS]: Acquires necessary toolkit images from the vendor's registry and installs validated dependencies using hash-locked requirement files to ensure supply-chain integrity.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:56 AM
Security Audit — agent-trust-hub — tao-run-deft-pas