home / skills / jeremylongshore / claude-code-plugins-plus-skills / adk-engineer

adk-engineer skill

safe

/plugins/ai-ml/jeremy-adk-software-engineer/skills/adk-engineer

This skill designs and implements production-ready ADK agents with structured code, tests, and deployment plans for reliable automation.

npx playbooks add skill jeremylongshore/claude-code-plugins-plus-skills --skill adk-engineer

Review the files below or copy the command above to add this skill to your agents.

Files (3)

SKILL.md

3.2 KB

---
name: adk-engineer
description: |
  Execute software engineer specializing in creating production-ready ADK agents with best practices, code structure, testing, and deployment automation. Use when asked to "build ADK agent", "create agent code", or "engineer ADK application". Trigger with relevant phrases based on skill purpose.
allowed-tools: Read, Write, Edit, Grep, Glob, Bash(cmd:*)
version: 1.0.0
author: Jeremy Longshore <[email protected]>
license: MIT
---
# ADK Engineer

Engineer production-ready Agent Development Kit (ADK) agents and multi-agent systems: clean structure, testability, safe tool usage, and deployment automation.

## Overview

Use this skill to design and implement ADK agent code that is maintainable and shippable: clear module boundaries, structured tool interfaces, regression tests, and a deployment checklist (local or Agent Engine).

## Prerequisites

- A target runtime (Python/Java/Go) consistent with the project’s pinned versions
- ADK installed (and any required model/provider SDKs configured)
- A test runner available in the repo (unit tests at minimum)
- If deploying: access to a Google Cloud project and permissions for the chosen deployment target

## Instructions

1. Clarify requirements: agent goals, tool surface, latency/cost constraints, and deployment target.
2. Propose architecture: single agent vs multi-agent, orchestration pattern, state strategy (Memory Bank / external store).
3. Scaffold structure: agent entrypoint(s), tool modules, config, and tests.
4. Implement incrementally:
   - add one tool at a time with input validation and structured outputs
   - add regression tests for each tool and critical prompt flows
5. Add operational guardrails: retries/backoff, timeouts, logging, and safe error messages.
6. Validate locally (tests + smoke prompts) and provide a deployment plan (when requested).

## Output

- A concrete architecture plan and file layout
- Agent and tool implementations (or patches) with tests
- A validation checklist (commands to run, expected outputs, and failure triage)
- Optional: deployment instructions and post-deploy health checks

## Error Handling

- Build/test failures: isolate the failing module, minimize the repro, fix, and add a regression test.
- Tool/runtime errors: enforce structured error responses and safe retries where appropriate.
- Deployment failures: provide the exact failing command, logs to inspect, and least-privilege IAM fixes.

## Examples

**Example: Productionizing an existing ADK agent**
- Request: “Refactor this agent into a clean module structure and add tests before we deploy.”
- Result: reorganized `src/` layout, tool boundaries, a test suite, and a deployment checklist.

**Example: Multi-agent workflow**
- Request: “Build a validator + deployer + monitor agent team with a sequential orchestrator.”
- Result: orchestrator skeleton, per-agent responsibilities, and smoke tests for each step.

## Resources

- Full detailed playbook (kept for reference): `{baseDir}/references/SKILL.full.md`
- Repo standards (source of truth):
  - `000-docs/6767-a-SPEC-DR-STND-claude-code-plugins-standard.md`
  - `000-docs/6767-b-SPEC-DR-STND-claude-skills-standard.md`
- ADK / Agent Engine docs: https://cloud.google.com/vertex-ai/docs/agent-engine

Overview

This skill engineers production-ready Agent Development Kit (ADK) agents and multi-agent systems with maintainability, testability, and deployment automation in mind. I deliver clear module boundaries, structured tool interfaces, regression tests, and an operational checklist for local or Agent Engine deployment. The outcome is shippable agent code that follows engineering best practices and reduces runtime surprises.

How this skill works

I start by clarifying agent goals, tool surfaces, latency/cost constraints, and the intended deployment target. Next I propose an architecture (single vs multi-agent), scaffold a clean repository layout, and implement tools incrementally with input validation and structured outputs. Each implementation is accompanied by unit/regression tests, operational guardrails (timeouts, retries, logging), and a validation checklist for local and cloud deployment.

When to use it

You need a production-quality ADK agent or multi-agent workflow ready for testing and deployment.
You want to refactor a prototype agent into a maintainable codebase with tests.
You need a scaffolded architecture and file layout before adding features.
You require deployment guidance for local environments or Google Agent Engine.
You want automated tests and operational guardrails to reduce on-call incidents.

Best practices

Design clear tool interfaces with explicit input/output schemas and validation.
Implement one tool at a time and add a regression test for each feature.
Keep agent entrypoints thin and move business logic into testable modules.
Add retries/backoff, request timeouts, structured logging, and safe error messages.
Include a validation checklist with reproducible commands, expected outputs, and triage steps.

Example use cases

Refactor a research agent into an ADK project: reorganize src/, add tool boundaries, and create unit tests.
Build a validator + deployer + monitor multi-agent team with a sequential orchestrator and smoke tests.
Scaffold a new agent: entrypoint, config, tools folder, test suite, and CI job template.
Hardening an existing agent by adding input validation, structured errors, and retry policies.
Prepare deployment to Agent Engine: deployment checklist, required IAM roles, and post-deploy health checks.

FAQ

What runtimes and tools are required?

Use a consistent pinned runtime (Python by default), ADK installed, model/provider SDKs configured, and a test runner available in the repo.

How do you handle deployment failures?

I provide the failing command, relevant logs to inspect, and least-privilege IAM fixes plus rollback or retry guidance.