setup-tangram-remote-tool

Pass

Audited by Gen Agent Trust Hub on Oct 11, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches requirements and setup configuration files from connect.aiscientist.tools and raw.githubusercontent.com/mims-harvard.
  • [REMOTE_CODE_EXECUTION]: The instructions include installing a Python wheel (tuplatform_connect) directly from an external URL. Although the installation command uses a SHA-256 hash fragment to verify file integrity, it still involves the execution of binary code from an external source.
  • [COMMAND_EXECUTION]: The skill utilizes several shell commands to manage virtual environments, install dependencies, and launch background services (tu serve).
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: The tool processes external genomic datasets (H5AD files) provided via file paths in the run_tangram_deconvolution operation.
  • Boundary markers: The documentation instructs users to verify scientific output and ensure no traces or secrets are present, though no automated boundaries are enforced in the scripts.
  • Capability inventory: The tu utility facilitates network binding and service sharing; the environment permits file system access for logging and caching.
  • Sanitization: The skill recommends running a preflight validation script (setup_skill_preflight.py) to verify the implementation before deployment.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 11, 2026, 05:27 PM