The Agent Skills Directory

[PERSISTENCE]: The skill provides instructions for persistent operation of the tool relay by using the tuplatform-service install command, which configures the relay as a background service or daemon.
[COMMAND_EXECUTION]: Users are instructed to run shell commands to verify their environment and tool configuration, such as checking CUDA/GPU availability via python -c and validating the MCP schema using an inline Python test script.
[EXTERNAL_DOWNLOADS]: The workflow requires the installation of external Python libraries (tooluniverse, fastmcp) and communication with a remote backend service hosted on the Render platform (tooluniverse-backend.onrender.com) to relay tool calls.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates hosting tools that process external data, creating a potential surface for indirect injection.
Ingestion points: Tools created with the @remote_tool decorator in my_tool.py accept caller-controlled inputs such as strings and floats.
Boundary markers: While explicit prompt delimiters are not provided in the code example, the documentation instructs users to manually validate input types and lengths.
Capability inventory: The skill facilitates the execution of local Python functions and the relaying of network traffic to a central platform.
Sanitization: The instructions proactively advise developers to sanitize error messages, use bounded inputs, and ensure no secrets or local filesystem paths are returned in tool responses.

host-and-share-remote-tool