Skills
skills/promptingcompany/nv-skills/tao-train-depth-anything-v2/Gen Agent Trust Hub

tao-train-depth-anything-v2

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches pre-trained backbone weights from a public asset storage service for model initialization.
  • Evidence: references/spec-param-inference.md contains a structured reference to download weights from https://dl.fbaipublicfiles.com/dinov2/dinov2_vitl14/dinov2_vitl14_pretrain.pth.
  • [COMMAND_EXECUTION]: Provides metadata and instructions for the agent to execute the depth_net CLI and Docker commands to manage the training and deployment lifecycle of AI models.
  • Evidence: references/skill_info.yaml defines several actions like train, evaluate, and inference that execute shell commands.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted dataset annotation files to be processed by CLI tools.
  • Ingestion points: Dataset annotation files (annotations.txt) processed in SKILL.md and references/spec-overrides.md.
  • Boundary markers: Absent.
  • Capability inventory: Shell command execution via the depth_net tool defined in references/skill_info.yaml.
  • Sanitization: Absent.
  • [SAFE]: No malicious behavior, obfuscation, or unauthorized access to sensitive system information was detected. All activities are well-documented and align with the primary purpose of AI model development.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 04:35 PM
Security Audit — agent-trust-hub — tao-train-depth-anything-v2