addon-google-agent-dev-kit（addon Google Agent DEV KIT 命令行）

Add-on: Google Agent Development Kit

Use this skill when a project needs Google Agent Development Kit based agent workflows and runtime contracts.

Compatibility

• Works with architect-python-uv-fastapi-sqlalchemy and architect-nextjs-bun-app.
• Can be combined with addon-langgraph-agent when graph-based execution is also required.
• Prefer Python-first implementation when both runtime options are viable.
• Can be combined with addon-llm-judge-evals; when used together, declare google-adk in config/skill_manifest.json so the judge runner can resolve the backend deterministically.

Inputs

Collect:

• ADK_RUNTIME: python | nextjs (default from selected architecture).
• ADK_DEFAULT_MODEL: default model id for ADK-backed runs.
• AGENT_GOAL: one-sentence scope statement.
• ALLOW_TOOLS: explicit list of allowed tools.
• MAX_AGENT_STEPS: default 10.
• TIMEOUT_SECONDS: default 60.

Integration Workflow

1. Add dependencies by runtime:

• Python:

uv add google-adk pydantic-settings

• Next.js:

# Use the project's package manager (examples):
bun add @google/adk zod
pnpm add @google/adk zod

1. Add files by runtime:

• Python:

src/{{MODULE_NAME}}/agent/adk_runtime.py
src/{{MODULE_NAME}}/agent/tools.py
src/{{MODULE_NAME}}/api/routes/agent.py

• Next.js:

src/lib/agent/adk-runtime.ts
src/lib/agent/tools.ts
src/app/api/agent/runs/route.ts

1. Enforce agent policy boundaries:

• Allow only explicitly approved tools.
• Add bounded step count and timeout.
• Emit run telemetry and decision traces.

1. Add fallback mode:

• If ADK provider/runtime is unavailable, return explicit degraded status and retry guidance.

1. If addon-llm-judge-evals is also selected:

• emit config/skill_manifest.json with addon-google-agent-dev-kit in addons
• declare "judge_backends": ["google-adk"] in capabilities
• allow the judge runner to reuse ADK_DEFAULT_MODEL when JUDGE_MODEL is unset

Required Template

Agent run status shape

{
  "runId": "uuid",
  "status": "queued|running|completed|failed|degraded",
  "steps": 0
}

Guardrails

• Documentation contract for generated code:

- Python: write module docstrings and docstrings for public classes, methods, and functions. - Next.js/TypeScript: write JSDoc for exported components, hooks, utilities, and route handlers. - Add concise rationale comments only for non-obvious logic, invariants, or safety constraints. - Apply this contract even when using template snippets below; expand templates as needed.

• Keep API credentials server-only.
• Do not expose unrestricted tool execution from public routes.
• Preserve auditability of agent decisions and failures.
• Keep degraded behavior explicit and non-silent.
• If judge evals are enabled, do not silently mix ADK and non-ADK judge implementations; require explicit backend selection when multiple judge-capable addons are present.

Validation Checklist

• Confirm generated code includes required docstrings/JSDoc and rationale comments for non-obvious logic.

uv run ruff check . || true
uv run mypy src || true
# Use the project's package manager (examples):
bun run lint || true
pnpm run lint || true
rg -n "degraded|ALLOW_TOOLS|MAX_AGENT_STEPS|status" src || true

• Manual checks:
• Agent run endpoint returns bounded status contract.
• Disallowed tool invocation fails with controlled error.

Decision Justification Rule

• Every non-trivial decision must include a concrete justification.
• Capture the alternatives considered and why they were rejected.
• State tradeoffs and residual risks for the chosen option.
• If justification is missing, treat the task as incomplete and surface it as a blocker.