The Agent Skills Directory

[DYNAMIC_EXECUTION]: In references/tool-builder.md, the skill provides an implementation example for a 'calculate' tool that uses the Python eval() function on strings generated by the AI model. Since AI-generated content can be influenced by untrusted external data or malicious prompts, this template encourages a pattern that allows an attacker to achieve arbitrary code execution on the system running the SDK.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data through automated file uploads (references/files.md) and web search tools (references/agent-patterns.md). The provided code examples for handling this data do not employ boundary markers, XML-style tagging, or input sanitization to differentiate between data and instructions. Combined with the skill's capabilities to execute local code and make authenticated network requests via webhooks, this creates an actionable surface for indirect prompt injection.
[EXTERNAL_DOWNLOADS]: The skill requires the inferencesh package from PyPI for its core functionality and references standard libraries including requests, aiohttp, and fastapi for extended features like streaming and web integration.

python-sdk