compute-env-setup

Warn

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to configure remote environments and employs 'sed' to modify Python source code files at runtime to fix version-specific compatibility issues in tools like 'ligandmpnn' and 'diffdock'.
  • [EXTERNAL_DOWNLOADS]: It fetches software components from remote sources, including binary tarballs for tools like 'bwa-mem2' and 'mmseqs' and models from various community repositories.
  • [REMOTE_CODE_EXECUTION]: The skill installs code directly from Git URLs and implements a library 'shim' strategy where a custom script ('torchtext_shim.py') is injected into the environment via a '.pth' file to override archived library behavior.
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface via the 'compute_details' tool.
  • Ingestion points: Environment configurations are read from persistent state stored in provider-specific ledger blocks.
  • Boundary markers: No delimiters or specific markers are used to separate configuration data from potential instructions within the markdown blocks.
  • Capability inventory: The skill can execute arbitrary shell commands via pip_phases and run_commands, and write files via shim_files.
  • Sanitization: No validation or sanitization of the configuration data retrieved from the provider is performed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 2, 2026, 01:21 AM
Security Audit — agent-trust-hub — compute-env-setup