The Agent Skills Directory

[Command Execution Capability]: The skill documentation includes details on integrating shell-based tools such as LocalShellTool and DockerShellTool. It provides necessary security context, noting that these tools should be sandboxed and that command filters alone do not constitute a complete security boundary. The host is instructed to own the sandboxing and executor lifecycle.
[Indirect Prompt Injection Surface]: As the framework is designed to process external and potentially untrusted data, it acknowledges the surface for indirect prompt injection. It suggests mitigation strategies such as using ContextProviders for dynamic instruction management and implementing Data Loss Prevention policies for binary content that cannot be easily distinguished from text.
[External Resource Integration]: The skill facilitates the retrieval of information from official documentation sources and code repositories. These integrations are performed through documented protocols (such as MCP) and are intended to keep the developer's environment aligned with the latest framework updates and samples.
[Sensitive Data Management]: Guidelines are provided for the secure handling of session data, conversation history, and authentication credentials. The skill emphasizes the importance of avoiding hardcoded secrets and using environment-based configuration management like load_dotenv().

agent-framework-builder-python