setup-liana-remote-tool
Warn
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Downloads and installs the 'tuplatform-connect' Python package from a remote URL ('https://connect.aiscientist.tools/downloads/tuplatform_connect-0.3.0-py3-none-any.whl'). While the installation includes a SHA-256 hash for integrity checking, the source is an external domain not recognized as a trusted vendor.
- [COMMAND_EXECUTION]: Requires the agent to run various shell commands for virtual environment creation, pip package management, and interacting with the 'tu' CLI for remote tool sharing and login.
- [DYNAMIC_EXECUTION]: Utilizes bash here-documents to execute dynamically generated Python scripts for checking the status of the tool's discovery and running example operations.
- [INDIRECT_PROMPT_INJECTION]: The skill's tools ingest external parameters such as file paths ('adata_path') and dataset keys ('cluster_key'), which constitutes an attack surface for indirect prompt injection. (1) Ingestion points: Arguments passed to the 'run_liana_cellphonedb' tool operation (SKILL.md). (2) Boundary markers: None identified. (3) Capability inventory: Execution of local command-line tools and shell-piped Python scripts (SKILL.md). (4) Sanitization: No explicit sanitization or input validation logic is described.
Audit Metadata