The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates and instructions for creating Python-based tools that execute local code. It describes a mechanism where Python scripts in the .tooluniverse/tools/ directory or within plugin packages are automatically imported and executed by the platform's discovery system.
[PROMPT_INJECTION]: The documented tool templates facilitate the ingestion of untrusted data from external APIs, creating a surface for indirect prompt injection if malicious instructions are returned in the API responses.
Ingestion points: API responses fetched using requests.get and urllib.request.urlopen in the my_tool.py and my_api_tool.py examples.
Boundary markers: No delimiters or instructions to ignore embedded commands are included in the tool templates.
Capability inventory: Tools possess network access (via requests and urllib) and the ability to execute logic defined in their run methods.
Sanitization: The provided code snippets do not include input validation or output sanitization to filter potentially malicious content from API responses.

tooluniverse-custom-tool