tao-train-depth-anything-v2

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches pretrained DINOv2 model weights from Meta's official public storage domain. This is an expected behavior for a machine learning skill to load necessary foundation model backbones.
  • [COMMAND_EXECUTION]: Generates and instructs the agent to execute docker run and depth_net CLI operations. These commands are fundamental to the skill's primary purpose of model training, evaluation, and inference.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface for indirect prompt injection as it processes user-provided datasets and configuration files.
  • Ingestion points: External data enters the agent context through user-supplied annotation files and configuration spec overrides defined in SKILL.md.
  • Boundary markers: Explicit boundary markers or warnings to ignore embedded instructions are absent in the documentation for user-provided data files.
  • Capability inventory: The skill leverages the Bash tool to execute docker and training commands across its scripts.
  • Sanitization: Input validation relies on the internal parsing and schema validation logic of the TAO Toolkit CLI tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:56 AM
Security Audit — agent-trust-hub — tao-train-depth-anything-v2