← ClaudeAtlas

google-adklisted

Develop Google Agent Development Kit applications with Agent, Runner, sessions, callbacks, evaluation, and Gemini models. Use for Python google-adk code.
fmind/dot · ★ 4 · AI & Automation · score 80
Install: claude install-skill fmind/dot
# Google ADK Implement Python ADK behavior within the existing project; [agents-cli](../agents-cli/SKILL.md) owns the generated application, evaluation commands, and deployment lifecycle. ## Workflow 1. **Inspect the installed SDK** and project model, tools, session service, and app entry point before coding. Use [python-stack](../python-stack/SKILL.md) for shared project defaults only when needed; preserve the generated layout. 1. **Choose the official guidance** below for ADK agents, tool functions, orchestration, callbacks, state, or tests; compare its supported SDK version with the project lock. 1. **Set the runtime contract**: choose one agent unless the workflow requires orchestration; make the model, provider authentication, session persistence, run limits, and timeout explicit. Keep the runner and session service consistent on app, user, and session IDs. 1. **Keep agent wiring explicit**: in the generated layout, `app/agent.py` defines `root_agent`; typed function signatures and docstrings describe its tools, while business logic stays in independently testable modules. Offline tests inspect agent wiring and call tool functions directly. 1. **Implement narrow tools** with complete types, useful docstrings, bounded I/O, and explicit errors. Keep business logic independently testable and prompts in reviewable source. Enforce authorization and idempotency in the tool boundary rather than relying on model instructions. 1. **Handle events and state**: consume the runner