google-adk-agents
Build, evaluate, and deploy agents with Google's Agent Development Kit (ADK). Use when you want code-first multi-agent systems, workflow agents, MCP tools, or Google-supported agent deployment paths.
SKILL.mdSkills/Agentic_AI/Google_ADK_Agents/SKILL.md
GitHubGoogle ADK Agents
Use this skill when you want Google's agent framework primitives instead of retrofitting agent behavior onto a generic app framework.
Workflow
- Pick the runtime language first: Python is the default, but ADK also supports TypeScript, Go, and Java.
- Decide whether the problem is best modeled as an LLM agent, a workflow agent, or a multi-agent system.
- Choose the tool surface: function tools, MCP tools, OpenAPI tools, or native integrations.
- Define evaluation and runtime strategy before deployment, not after.
- Validate locally, then promote to managed runtime or production infra only after passing a small eval set.
Guardrails
- Keep agent boundaries explicit; avoid one giant catch-all agent.
- Prefer documented ADK primitives over undocumented framework internals.
- Record the model/provider abstraction used so the workflow stays reproducible.
- Treat third-party integrations as optional until auth, quotas, and failure modes are documented.
Output Requirements
- State the ADK runtime and language.
- State the agent topology (single, workflow, or multi-agent).
- State the evaluation plan and one production risk.