setup-singler-remote-tool

Warn

Audited by Gen Agent Trust Hub on Oct 11, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads requirements files and software packages from connect.aiscientist.tools and GitHub. While the GitHub resources belong to the skill author's organization, the aiscientist.tools domain is an external source.
  • [REMOTE_CODE_EXECUTION]: The instructions include installing a Python wheel directly from a URL (tuplatform-connect from connect.aiscientist.tools). Although the download is pinned with a SHA-256 hash, installing code from unverified external domains carries inherent risk.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands to manage environments and executes R scripts (Rscript) to install Bioconductor packages.
  • [DATA_EXFILTRATION]: The skill connects to a remote relay service located at tooluniverse-backend.onrender.com, establishing a network path to an external service.
  • [INDIRECT_PROMPT_INJECTION]: The run_singler_annotate tool ingests HDF5 data files (.h5ad), which creates a surface for indirect prompt injection if the output is not properly sanitized. 1. Ingestion points: The adata_path parameter in SKILL.md allows processing of external data files. 2. Boundary markers: No delimiters or instructions are present to prevent the model from obeying instructions embedded within the data. 3. Capability inventory: The skill can execute shell commands and tool calls. 4. Sanitization: No sanitization or validation of the data file contents is described.
  • [DYNAMIC_EXECUTION]: The skill performs runtime installation of Python and R dependencies based on remote configuration files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 11, 2026, 05:27 PM