tao-analyze-gaps-visual-changenet

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes docker run commands to perform data analysis. It uses host path bind-mounts (-v $WORKSPACE:$WORKSPACE) to provide the container with access to local experiment data and to save outputs back to the workspace. It also utilizes a secondary alpine container to perform chown operations on the generated results.
  • [EXTERNAL_DOWNLOADS]: The skill pulls a Docker image from a remote registry. The image identifier (tao_toolkit.data_services) is retrieved from a local versions.yaml file, and the image is fetched using docker pull prior to execution.
  • [REMOTE_CODE_EXECUTION]: The skill runs containerized code to perform complex computational tasks, including decision threshold optimization and weakness ranking. This execution is performed with GPU acceleration (--gpus all) and is the primary purpose of the skill.
  • [DATA_EXFILTRATION]: A localized packaging script (hooks/rca-package.sh) copies session transcripts and skill configuration files into the analysis results directory. This provides a record of the analysis session for auditing purposes but involves reading metadata and settings from the environment's configuration directories.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 05:48 PM
Security Audit — agent-trust-hub — tao-analyze-gaps-visual-changenet