What problem does it solve? Building production agents with Google's Agent Development Kit involves many phases — design, scaffolding, coding, evaluation, deployment, publishing, and observability — and skipping steps leads to missing eval boilerplate, broken CI/CD, and untested behavior. This Skill orchestrates that entire workflow through the agents-cli toolkit so nothing critical is skipped. ## Core Features & Use Cases - Phased workflow orchestration: Walks through Phase 0 (design dialogue and spec) through Phase 7 (observability), cross-referencing the right sub-skill at each phase. - Code preservation and model selection rules: Enforces surgical code edits, prevents unwanted model changes, and provides commands to list current Gemini models. - Evaluation discipline: Distinguishes pytest (code correctness) from agents-cli eval (agent behavior), and forbids behavioral assertions on non-deterministic LLM output. - Use Case: A developer asks to build a customer-support agent with memory and guardrails. The Skill drives a design dialogue, writes .agents-cli-spec.md, points to matching clone-and-study recipes, scaffolds the project, and mandates an eval loop before any deployment. ## Quick Start Ask the assistant to develop an agent using ADK, for example: help me build and deploy an ADK agent that answers questions over our documentation.
