The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install a Python package from a remote URL outside of standard registries like PyPI. Specifically, it uses pip install to fetch tuplatform-connect from https://connect.aiscientist.tools/downloads/tuplatform_connect-0.3.0-py3-none-any.whl.
[REMOTE_CODE_EXECUTION]: The installation of the tuplatform-connect wheel from an external, non-standard domain constitutes remote code execution. Although the installation command includes a SHA-256 hash (3fad5eee5ecf7887a693d93ccd1aa112dc0955617a885d1fc3daded0030f9ae0) to ensure integrity, the source aiscientist.tools is not a recognized trusted vendor or well-known service.
[COMMAND_EXECUTION]: The skill makes extensive use of shell commands and custom CLI tools, including:
A local preflight script: python scripts/remote_validation/setup_skill_preflight.py.
The tu CLI tool for service management: tu remote login, tu remote share, and tu serve.
Inline Python scripts for validating MCP (Model Context Protocol) endpoints.
[INDIRECT_PROMPT_INJECTION]: The skill sets up an MCP server (boltz2_docking) that ingests and processes external biological data, which could serve as a vector for indirect injection.
Ingestion points: The sequence and ligands (SMILES strings) arguments in the boltz2_docking tool.
Boundary markers: None explicitly implemented in the provided documentation or snippets to delimit the scientific data from instructions.
Capability inventory: The skill manages local file system directories (mkdir -p caches/boltz runs/boltz) and performs network operations via the tu relay and fastmcp client/server architecture.
Sanitization: The instructions include recommendations to "Check scientific meaning, output bounds, invalid-input behavior, and absence of paths, secrets, and traces" in the tool's output.

setup-boltz-remote-tool