The Agent Skills Directory

Remote Installation Script: The documentation includes a command to install the uv tool via a script hosted on astral.sh (curl -LsSf https://astral.sh/uv/install.sh | sh). This is a common installation method for this tool from a well-known provider.
Filesystem and Artifact Middleware: The skill describes the use of Filesystem and Artifacts middleware, which can grant an agent the ability to read and write files within a designated directory (e.g., Filesystem(root_dir='./workspace', allow_write_access=True)). While these are intended features for building coding or research agents, they should be used with appropriate human-in-the-loop (HITL) approvals when processing untrusted input.
Indirect Prompt Injection Surface: As this skill teaches how to build agents that process user-supplied input and execute tools, it describes an architecture that is susceptible to indirect prompt injection. The skill correctly suggests using ToolApproval middleware to mitigate this by requiring human confirmation for sensitive operations.
Persistence Mechanisms: The documentation mentions FileSessionStore, which persists agent session data and history to the local filesystem (e.g., in a ./.snapshots directory). This is an intended persistence feature of the Genkit framework.
Command Execution Patterns: The skill describes the use of genkit start, which wraps other shell commands (like uv run) to provide telemetry and tracing. This is a core feature of the Genkit developer experience.
Development Environment Configuration: The FastAPI examples include a broad Cross-Origin Resource Sharing (CORS) policy (allow_origins=['*']). This is a common configuration for local development and testing but should be restricted in production environments.

developing-genkit-python