The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the uv package manager and ollama vision models using standard installation scripts from their official domains (astral.sh and ollama.com). These are well-known services in the developer community.
[CREDENTIALS_UNSAFE]: The skill demonstrates a strong security posture regarding authentication. It includes a dedicated project_setup.md contract that mandates the use of .gitignore to prevent secret leaks and provides a .env template. Furthermore, it recommends using platform-specific deny rules (e.g., .claude/settings.json) to restrict the agent from reading local environment files containing API keys.
[REMOTE_CODE_EXECUTION]: While the documentation includes shell commands to install tools via piped execution (e.g., curl | sh), these are directed at reputable, well-known service providers (Astral and Ollama) for environment setup and are standard practice for these tools.
[COMMAND_EXECUTION]: The skill facilitates command execution through the RouterAgent and FastAPI templates. These implementations are standard for tool-calling agents and web backends, with appropriate use of type annotations and structured models to scope execution.
[INDIRECT_PROMPT_INJECTION]: As a framework for RAG (Retrieval-Augmented Generation) systems, the skill acknowledges the risks of processing external data. It includes mitigation strategies such as specific prompt instructions for the LLM to adhere strictly to provided context and the use of Pydantic for input validation in API endpoints.

weaviate-cookbooks