google/adk-python
6 skills6 installs120k starsGitHub
Install
npx skills add https://github.com/google/adk-pythonSkills in this repo
1Adk Agent Builderadk-agent-builder is a reference skill for solo and indie builders who ship agents with Google ADK for Python and need production-grade workflow graphs instead of flat step lists. It documents advanced patterns: nested workflows that compose inner pipelines as nodes, RetryConfig and timeout wiring on nodes, custom node types with the required overrides, and dynamic execution using run_node with explicit rules about waiting states. The included verification checklist maps directly to ADK validation expectations so you catch structural mistakes—such as invalid cycles—before you integrate the graph into your app. Use it while you are designing multi-step agent pipelines, debugging resume/rerun behavior, or hardening orchestration after a basic ADK prototype works. It does not replace ADK installation or foundational agent tutorials; it accelerates the move from a single agent to a validated, composable workflow architecture you can maintain in your repo.1installs2Adk ArchitectureADK Architecture is an agent skill for solo builders shipping agents on Google’s Agent Development Kit for Python who must get human-in-the-loop and long-running tool flows right the first time. It explains how interrupts surface through ancestor nodes, which events get persisted at the leaf, and how the Runner resumes execution by scanning session history for the correct invocation_id. You learn why resumed nodes keep the same run_id while fresh dispatches get a new one, and how rerun_on_resume governs partial versus complete FunctionResponses on workflow-style nodes. That depth matters when your product pauses for approval, external API callbacks, or operator input without corrupting graph state. Use it while designing multi-node workflows, debugging stuck WAITING states, or reviewing agent PRs before Ship. It complements hands-on ADK coding skills by anchoring architecture rules that are easy to misread from scattered docs.1installs3Adk Debugadk-debug is a Google ADK Python skill that walks agents through systematic debugging of Agent Development Kit agents—whether you need the browser UI and REST surface of adk web or the faster CLI loop of adk run. Solo builders shipping ADK-based agents with Claude Code, Codex, or Windsurf invoke it when sessions misbehave, tool calls fail, event ordering looks wrong, or model responses need isolation from infrastructure noise. The skill encodes a clear preference order: default to adk run (and query mode inside it) for most development, reserve adk web for visualization, multi-turn UI testing, or API-server problems, and always check for an existing server with a health curl before spawning another background process. It covers starting verbose or hot-reload servers, run_in_background etiquette, and shutting down servers the agent started. Treat it as the first structured stop after reproducing a bug in an ADK agent, spanning Build-heavy agent-tooling work and Operate-style incident diagnosis when production-like sessions break.1installs4Adk Sample Creatoradk-sample-creator is an agent skill for solo builders and contributors who extend Google’s ADK Python repository with clear, reviewable examples. It walks you through picking the right subdirectory under contributing/, naming the sample in snake_case, and producing the minimal pair of agent.py and README.md that reviewers expect. The skill encodes ADK 2.0 conventions such as absolute imports for testability and leaving the model parameter unset on Agent instances so samples respect system-default models unless you explicitly need otherwise. Invoke it when you are adding demonstrations for dynamic nodes, fan-out and fan-in workflows, or standalone agents—not when you are shipping a private production service. Tip flow suggests reading adk-style first so architecture choices match current ADK guidance, which reduces rework during upstream PR review.1installs5Adk Setupadk-setup is a Google ADK Python onboarding skill that walks you through a reproducible local dev environment using Astral uv rather than ad-hoc pip workflows. Solo contributors and indie teams cloning adk-python use it on day one when they want dependencies, extras, linters, and test runners aligned with how maintainers expect patches. The skill sequences version checks, virtualenv creation with Python 3.11, full extra sync, pre-commit hook installation, tox with tox-uv, and the addlicense tool when Go is present. It is marked disable-model-invocation because setup commands should run deliberately in your shell. After verification tests pass, you are ready to implement agents, run examples, and open PRs without fighting dependency drift. Skip it if you only consume ADK as a published library in another app without hacking the framework repo.1installs6Adk Styleadk-style is a contributor-oriented agent skill that encodes how Google Agent Development Kit (ADK) Python code should read, organize, and test. It separates public API documentation expectations from lighter internal commentary, defines physical file boundaries in `workflow/` and private `_` modules, and maps every source path to a mirrored unittest path with optional multi-file test prefixes. Solo builders shipping custom ADK agents or patching toolkit behavior can invoke it whenever generated or hand-written Python would otherwise drift from upstream merge requirements. Following it reduces review churn, keeps imports ordered consistently, and makes agent tool modules discoverable through package `__init__.py` exports—especially important when multiple skills or tools land in the same repository over time.1installs