The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download a Python wheel (tuplatform-connect) from an external domain (https://connect.aiscientist.tools). While the instruction includes a SHA-256 hash for integrity verification, the domain is not a recognized trusted organization or well-known service provider.
[COMMAND_EXECUTION]: The instructions require the execution of numerous shell commands, including environment setup (venv), package installation (pip), and tool management via a custom CLI (tu), which modifies the local system state.
[PERSISTENCE]: The skill references management of a background service (tuplatform-service uninstall), which implies the installation process establishes a persistent service or daemon on the host machine to maintain the remote tool's availability.
[DATA_EXFILTRATION]: The core functionality involves "sharing" local tools via a relay service (tu remote share). This establishes a network path for external entities to interact with local tool endpoints, creating an inherent risk of data exposure depending on the tool's implementation.
[DYNAMIC_EXECUTION]: The skill provides verification examples that pipe multi-line Python code directly into the interpreter (`python
<<'PY'`), which is a form of dynamic script execution for local validation.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The run_scanvi_annotate tool processes external data files provided via the adata_path parameter (SKILL.md).
Boundary markers: The instructions do not define any delimiters or safety markers to isolate the data content from the agent's command context (SKILL.md).
Capability inventory: The skill provides capabilities for network relay, local server hosting, and background service management (SKILL.md).
Sanitization: No sanitization or validation logic is specified for the processed data files (SKILL.md).

setup-scanvi-remote-tool