The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building agents and flows that process untrusted user input through functions like ai.generate, chat.send, and agent.chat.\n
Ingestion points: User prompts and messages are ingested in SKILL.md, references/agents.md, and references/fastapi.md.\n
Boundary markers: The documentation mentions using .prompt files with structured YAML frontmatter and Handlebars templates (documented in references/dotprompt.md), which helps separate instructions from data.\n
Capability inventory: The documented agents can call tools, access the local filesystem via Filesystem middleware (documented in references/agents.md), and perform network requests through custom tools.\n
Sanitization: The documentation highlights ToolApproval middleware and human-in-the-loop (HITL) patterns in references/agents-human-in-the-loop.md as mechanisms to review and approve sensitive actions.\n- [EXTERNAL_DOWNLOADS]: The documentation instructs users to install the Genkit CLI and Python packages from well-known registries.\n
Evidence: npm install -g genkit-cli (SKILL.md) and uv add genkit (references/setup.md).\n- [COMMAND_EXECUTION]: The skill provides numerous shell command examples for project initialization, running development servers, and interacting with the Genkit CLI.\n
Evidence: genkit start -- uv run src/main.py (SKILL.md), export GEMINI_API_KEY=your-api-key-here (references/dev-workflow.md).

developing-genkit-python