setup-ldsc-remote-tool

Pass

Audited by Gen Agent Trust Hub on Oct 11, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches configuration and dependency lists from the author's official GitHub repository (mims-harvard/ToolUniverse) and the project's infrastructure domain (aiscientist.tools).
  • [EXTERNAL_DOWNLOADS]: Downloads a Python wheel (tuplatform-connect) from a remote server for installation. The download includes a pinned SHA-256 hash to ensure the integrity of the binary file.
  • [COMMAND_EXECUTION]: Provides instructions to execute shell commands for environment creation, dependency installation, and service management (uv, pip, tu serve).
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of a tool interface (run_ldsc_heritability) that ingests external data files, creating a potential surface for indirect injection.
  • Ingestion points: The sumstats_path parameter in the run_ldsc_heritability tool allows the agent to process external files (SKILL.md).
  • Capability inventory: The environment enables file operations, network relaying via ToolUniverse Connect, and Python code execution (SKILL.md).
  • Boundary markers: The documentation includes notes to verify output for the absence of secrets, paths, and traces.
  • Sanitization: Instructions do not explicitly detail input sanitization, which is expected to be handled by the underlying tool implementation.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 11, 2026, 05:27 PM