tao-train-reid

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill defines several shell commands in the references/skill_info.yaml file, such as 're_identification train -e {config_path}', to be executed via the Bash tool. These commands are the primary intended functionality and are constructed using configuration files validated against provided schemas.
  • [DATA_EXFILTRATION]: Configuration templates, including references/spec_template_train.yaml, enable logging to Weights & Biases (wandb) by default. This transmits training metadata and metrics to the well-known wandb.ai service.
  • [SAFE]: The skill processes user-provided dataset paths and specification overrides as documented in SKILL.md.
  • Ingestion points: Dataset paths (S3 or local) and spec_overrides supplied by the user or upstream workflows.
  • Boundary markers: The skill relies on structured configuration files and JSON schemas, such as schemas/train.schema.json, which define valid input structures and data boundaries.
  • Capability inventory: The skill is limited to executing shell commands for the re_identification CLI tool within the TAO Toolkit container environment.
  • Sanitization: Input values are validated against the provided JSON schemas and the TAO Toolkit's internal configuration parser.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 04:39 PM
Security Audit — agent-trust-hub — tao-train-reid