[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7574B94PFXYK9BV6VMY7WQ7":3,"skill-same:01M7574B94PFXYK9BV6VMY7WQ7":59},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":37,"commitSha":38,"skillDirectory":39,"forksCount":14,"repositoryHtmlUrl":40,"repositoryHomepageUrl":6,"ownerHtmlUrl":41,"ownerBlogUrl":42,"ownerDescription":43,"updatedAt":44,"install":45,"resources":58},"01M7574B94PFXYK9BV6VMY7WQ7","google-adk",null,"Use when building multi-agent systems with Google's Agent Development Kit (ADK), configuring agents, tools, workflows, state, and sessions.","SatangThevalue\u002Fai-skills","SatangThevalue","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F17944930?v=4","01M756VHQQ58XSM5PDS03W79VY","Sa-Tang","01M756VHRNYXRY7HM230S4YHVV",0,"MIT","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautonomous-ai-agents\u002Fgoogle-adk\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: google-adk\ndescription: Use when building multi-agent systems with Google's Agent Development Kit (ADK), configuring agents, tools, workflows, state, and sessions.\nversion: 1.0.0\nauthor: Hermes Agent\nlicense: MIT\nmetadata:\n  hermes:\n    tags: [google-adk, agents, multi-agent, framework, gemini, google-cloud]\n    related_skills: [agent-frameworks-integration, plan]\n---\n\n# Google ADK (Agent Development Kit) Workflow\n\n## Overview\nGoogle's Agent Development Kit (ADK) is an open-source, code-first framework designed for building, evaluating, and deploying multi-agent systems. It treats agents as structured software components rather than simple prompt chains. It features first-class orchestration (hierarchical\u002Fdynamic), native Agent2Agent (A2A) protocol support, structured context management, and deep observability via ADK Studio.\n\n## When to Use\n- Building multi-agent workflows requiring sequential, parallel, or loop structures.\n- Integrating Gemini models (or other models via LiteLLM) in a code-first, structured environment.\n- Requiring deep tracing of agent decisions, prompts, state updates, and latency (via ADK Studio).\n- Implementing Agent-to-Agent (A2A) communication across different frameworks.\n\n### Do Not Use For:\n- Simple, single-prompt chat scripts with no routing or state requirements.\n- Environments where installing the ADK Python\u002FJava\u002FGo\u002FTypeScript SDK is restricted.\n\n## Setup & Preparation\n\n### 1. Installation\nInstall the Google ADK Python SDK (typically done within a virtual environment):\n```bash\npip install google-adk\n```\nEnsure you have the required API key for Gemini set in your environment (e.g., `GEMINI_API_KEY`) or GCP Vertex AI authenticated.\n\n### 2. Project Structure\nADK prefers a structured layout. Create a clean project module:\n```\nmy_agent_project\u002F\n├── main.py\n├── agents.py\n├── tools.py\n└── README.md\n```\n\n## Core Implementation Patterns\n\n### 1. Defining Agents and Tools\nAgents are instances of `LlmAgent`. Tools are normal Python functions that accept a `ToolContext` argument.\n\n```python\nfrom google.adk.agents import LlmAgent\nfrom google.adk.tools import ToolContext\n\n# Tools use docstrings for LLM selection\ndef exit_loop(tool_context: ToolContext) -> dict:\n    \"\"\"Call this when the draft is good enough to publish.\"\"\"\n    tool_context.actions.escalate = True\n    tool_context.actions.skip_summarization = True\n    return {}\n\nwriter_agent = LlmAgent(\n    name=\"writer_agent\",\n    model=\"gemini-2.5-flash\",\n    instruction=\"You write short popular-science articles. Write about 200 words.\",\n    output_key=\"draft\",\n)\n\ncritic_agent = LlmAgent(\n    name=\"critic_agent\",\n    model=\"gemini-2.5-flash\",\n    instruction=\"You are a sharp editor. Review the draft. If good, call exit_loop.\",\n    tools=[exit_loop],\n    output_key=\"critique\",\n)\n```\n\n### 2. Deterministic Orchestration\nUse structured wrapper agents to control flow:\n- `SequentialAgent`: Runs sub-agents in a fixed sequence.\n- `ParallelAgent`: Runs sub-agents concurrently.\n- `LoopAgent`: Runs sub-agents iteratively up to `max_iterations` or until a tool escalates.\n\n```python\nfrom google.adk.agents import LoopAgent, SequentialAgent\n\nrefinement_loop = LoopAgent(\n    name=\"refinement_loop\",\n    sub_agents=[writer_agent, critic_agent],\n    max_iterations=3,\n)\n\nroot_agent = SequentialAgent(\n    name=\"writing_pipeline\",\n    sub_agents=[theme_agent, refinement_loop],\n)\n```\n\n### 3. State Management & Instruction Templating\nState is a key-value store shared across a session. Agents write to the key specified in `output_key`. Other agents read this data using curly braces `{key}`:\n- `{topic}`: Required key. If not present in state, execution will error out.\n- `{critique?}`: Optional key. Evaluates to empty string if not present in state.\n\n```python\n# In writer_agent instruction:\n# \"Write a short article about {topic}. Review previous feedback: {critique?}\"\n```\n\n### 4. Runner & Session Service\nTo run the agent, initialize a `Runner` with a `SessionService` (like `InMemorySessionService` for local dev):\n\n```python\nimport asyncio\nfrom google.adk.sessions import InMemorySessionService\nfrom google.adk.runners import Runner\n\nasync def main():\n    session_service = InMemorySessionService()\n    runner = Runner(\n        agent=root_agent,\n        app_name=\"writing_assistant\",\n        session_service=session_service,\n    )\n    \n    session_id = \"session_1\"\n    user_id = \"user_123\"\n    \n    await session_service.create_session(\n        app_name=\"writing_assistant\",\n        user_id=user_id,\n        session_id=session_id,\n    )\n    \n    async for event in runner.run_async(\n        user_id=user_id,\n        session_id=session_id,\n        new_message=\"Let's write about quantum computing.\",\n    ):\n        print(f\"Event: {event}\")\n        \n    # Extract output state\n    session = await session_service.get_session(\n        app_name=\"writing_assistant\",\n        user_id=user_id,\n        session_id=session_id,\n    )\n    print(\"Final Draft:\", session.state.get(\"draft\"))\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\n### 5. Memory Management (Cross-Session)\nArchiving finished session state to long-term memory:\n```python\nasync def archive_run(callback_context):\n    await callback_context.add_session_to_memory()\n    return None\n```\nRetrieving memory from a tool:\n```python\nfrom google.adk.tools import load_memory\n# Pass load_memory in tools list and instruct the agent to use it\n```\n\n## Running ADK Studio\nTo observe execution paths, prompt history, state transitions, and latency, launch ADK Studio in your project root:\n```bash\nadk web\n```\nThis starts a local dashboard where you can step through Events and inspect the Traces\u002FGraph.\n\n## Common Pitfalls\n1. **Model tool-calling docs:** ADK agents select tools based on docstrings. If your docstrings are ambiguous or missing, the model will call tools incorrectly.\n2. **Infinite loops:** If `LoopAgent` does not hit the stop condition, it runs up to `max_iterations`. Always set a safe limit for `max_iterations` and implement a clear exit tool using `tool_context.actions.escalate = True`.\n3. **Template errors:** Using `{key}` in instruction templates raises errors if the key is not initialized in the state yet. Use `{key?}` for optional keys.\n4. **Skip Summarization:** For exit or utility tools that output nothing, set `tool_context.actions.skip_summarization = True` to avoid redundant LLM calls.\n\n## Verification Checklist\n- [ ] ADK SDK installed (`google-adk`) and python import works.\n- [ ] API keys (`GEMINI_API_KEY` etc.) exported.\n- [ ] Exit tools explicitly set `tool_context.actions.escalate = True`.\n- [ ] Optional state references in instructions formatted with `?` (e.g. `{critique?}`).\n- [ ] Local dashboard verified via `adk web` during trial runs.\n","# Google ADK (Agent Development Kit) Workflow\n\n## Overview\nGoogle's Agent Development Kit (ADK) is an open-source, code-first framework designed for building, evaluating, and deploying multi-agent systems. It treats agents as structured software components rather than simple prompt chains. It features first-class orchestration (hierarchical\u002Fdynamic), native Agent2Agent (A2A) protocol support, structured context management, and deep observability via ADK Studio.\n\n## When to Use\n- Building multi-agent workflows requiring sequential, parallel, or loop structures.\n- Integrating Gemini models (or other models via LiteLLM) in a code-first, structured environment.\n- Requiring deep tracing of agent decisions, prompts, state updates, and latency (via ADK Studio).\n- Implementing Agent-to-Agent (A2A) communication across different frameworks.\n\n### Do Not Use For:\n- Simple, single-prompt chat scripts with no routing or state requirements.\n- Environments where installing the ADK Python\u002FJava\u002FGo\u002FTypeScript SDK is restricted.\n\n## Setup & Preparation\n\n### 1. Installation\nInstall the Google ADK Python SDK (typically done within a virtual environment):\n```bash\npip install google-adk\n```\nEnsure you have the required API key for Gemini set in your environment (e.g., `GEMINI_API_KEY`) or GCP Vertex AI authenticated.\n\n### 2. Project Structure\nADK prefers a structured layout. Create a clean project module:\n```\nmy_agent_project\u002F\n├── main.py\n├── agents.py\n├── tools.py\n└── README.md\n```\n\n## Core Implementation Patterns\n\n### 1. Defining Agents and Tools\nAgents are instances of `LlmAgent`. Tools are normal Python functions that accept a `ToolContext` argument.\n\n```python\nfrom google.adk.agents import LlmAgent\nfrom google.adk.tools import ToolContext\n\n# Tools use docstrings for LLM selection\ndef exit_loop(tool_context: ToolContext) -> dict:\n    \"\"\"Call this when the draft is good enough to publish.\"\"\"\n    tool_context.actions.escalate = True\n    tool_context.actions.skip_summarization = True\n    return {}\n\nwriter_agent = LlmAgent(\n    name=\"writer_agent\",\n    model=\"gemini-2.5-flash\",\n    instruction=\"You write short popular-science articles. Write about 200 words.\",\n    output_key=\"draft\",\n)\n\ncritic_agent = LlmAgent(\n    name=\"critic_agent\",\n    model=\"gemini-2.5-flash\",\n    instruction=\"You are a sharp editor. Review the draft. If good, call exit_loop.\",\n    tools=[exit_loop],\n    output_key=\"critique\",\n)\n```\n\n### 2. Deterministic Orchestration\nUse structured wrapper agents to control flow:\n- `SequentialAgent`: Runs sub-agents in a fixed sequence.\n- `ParallelAgent`: Runs sub-agents concurrently.\n- `LoopAgent`: Runs sub-agents iteratively up to `max_iterations` or until a tool escalates.\n\n```python\nfrom google.adk.agents import LoopAgent, SequentialAgent\n\nrefinement_loop = LoopAgent(\n    name=\"refinement_loop\",\n    sub_agents=[writer_agent, critic_agent],\n    max_iterations=3,\n)\n\nroot_agent = SequentialAgent(\n    name=\"writing_pipeline\",\n    sub_agents=[theme_agent, refinement_loop],\n)\n```\n\n### 3. State Management & Instruction Templating\nState is a key-value store shared across a session. Agents write to the key specified in `output_key`. Other agents read this data using curly braces `{key}`:\n- `{topic}`: Required key. If not present in state, execution will error out.\n- `{critique?}`: Optional key. Evaluates to empty string if not present in state.\n\n```python\n# In writer_agent instruction:\n# \"Write a short article about {topic}. Review previous feedback: {critique?}\"\n```\n\n### 4. Runner & Session Service\nTo run the agent, initialize a `Runner` with a `SessionService` (like `InMemorySessionService` for local dev):\n\n```python\nimport asyncio\nfrom google.adk.sessions import InMemorySessionService\nfrom google.adk.runners import Runner\n\nasync def main():\n    session_service = InMemorySessionService()\n    runner = Runner(\n        agent=root_agent,\n        app_name=\"writing_assistant\",\n        session_service=session_service,\n    )\n    \n    session_id = \"session_1\"\n    user_id = \"user_123\"\n    \n    await session_service.create_session(\n        app_name=\"writing_assistant\",\n        user_id=user_id,\n        session_id=session_id,\n    )\n    \n    async for event in runner.run_async(\n        user_id=user_id,\n        session_id=session_id,\n        new_message=\"Let's write about quantum computing.\",\n    ):\n        print(f\"Event: {event}\")\n        \n    # Extract output state\n    session = await session_service.get_session(\n        app_name=\"writing_assistant\",\n        user_id=user_id,\n        session_id=session_id,\n    )\n    print(\"Final Draft:\", session.state.get(\"draft\"))\n\nif __name__ == \"__main__\":\n    asyncio.run(main())\n```\n\n### 5. Memory Management (Cross-Session)\nArchiving finished session state to long-term memory:\n```python\nasync def archive_run(callback_context):\n    await callback_context.add_session_to_memory()\n    return None\n```\nRetrieving memory from a tool:\n```python\nfrom google.adk.tools import load_memory\n# Pass load_memory in tools list and instruct the agent to use it\n```\n\n## Running ADK Studio\nTo observe execution paths, prompt history, state transitions, and latency, launch ADK Studio in your project root:\n```bash\nadk web\n```\nThis starts a local dashboard where you can step through Events and inspect the Traces\u002FGraph.\n\n## Common Pitfalls\n1. **Model tool-calling docs:** ADK agents select tools based on docstrings. If your docstrings are ambiguous or missing, the model will call tools incorrectly.\n2. **Infinite loops:** If `LoopAgent` does not hit the stop condition, it runs up to `max_iterations`. Always set a safe limit for `max_iterations` and implement a clear exit tool using `tool_context.actions.escalate = True`.\n3. **Template errors:** Using `{key}` in instruction templates raises errors if the key is not initialized in the state yet. Use `{key?}` for optional keys.\n4. **Skip Summarization:** For exit or utility tools that output nothing, set `tool_context.actions.skip_summarization = True` to avoid redundant LLM calls.\n\n## Verification Checklist\n- [ ] ADK SDK installed (`google-adk`) and python import works.\n- [ ] API keys (`GEMINI_API_KEY` etc.) exported.\n- [ ] Exit tools explicitly set `tool_context.actions.escalate = True`.\n- [ ] Optional state references in instructions formatted with `?` (e.g. `{critique?}`).\n- [ ] Local dashboard verified via `adk web` during trial runs.\n",{"name":5,"author":24,"license":15,"version":25,"metadata":26,"description":7},"Hermes Agent","1.0.0",{"hermes":27},{"tags":28,"related_skills":34},[5,29,30,31,32,33],"agents","multi-agent","framework","gemini","google-cloud",[35,36],"agent-frameworks-integration","plan","main","236825fbf4e5956a8cd2b0bf843494f0a5dfaf04","skills\u002Fautonomous-ai-agents\u002Fgoogle-adk","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills","https:\u002F\u002Fgithub.com\u002FSatangThevalue","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fsatangthevalue\u002F","Hi, I'm Sa-Tang. 🖐️🖐️🖐️\r\nThanks for following.\r\nสวัสดีครับ สตังค์ครับผม\r\nขอบคุณหรับการติดตามครับ","2026-10-10 12:10:40",{"source":8,"skillName":5,"commands":46,"prompt":56,"zipAvailable":57},[47,50,53],{"tool":48,"command":49},"npx","npx skills add SatangThevalue\u002Fai-skills --skill google-adk",{"tool":51,"command":52},"bunx","bunx skills add SatangThevalue\u002Fai-skills --skill google-adk",{"tool":54,"command":55},"pnpm","pnpm dlx skills add SatangThevalue\u002Fai-skills --skill google-adk","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fautonomous-ai-agents\u002Fgoogle-adk\u002FSKILL.md",true,[],[60,67,74,81,88,94],{"skillNo":61,"name":62,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M76TQY0W54NCMRVE9HB8MMTA","writing-plans","Write implementation plans: bite-sized tasks, paths, code.","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fwriting-plans\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M76TNN9AYY91KD4DT415K1FF","subagent-driven-development","Execute plans via delegate_task subagents (2-stage review).","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fsubagent-driven-development\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M76TME6SXP9SYJZCXX5FFZHR","react-composition-2026","Teaches modern React composition patterns for 2025\u002F2026. Use when designing component APIs, building shared UI libraries, or refactoring prop-heavy components.","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Freact-composition-2026\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M76TMAR2D5G9BKZWGEFJKWT2","react-2026","Provides a comprehensive guide to the modern React 2026 stack. Use when starting a new React project or modernizing an existing one with current frameworks, build tools, routing, state management, or AI integration.","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Freact-2026\u002FSKILL.md",[],[],{"skillNo":89,"name":36,"nameZh":6,"description":90,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":91,"categories":92,"keywords":93,"publishedAt":19},"01M76TM2FA8ZFZR1S6ZM34WFFK","Plan mode: write markdown plan to .hermes\u002Fplans\u002F, no exec.","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fplan\u002FSKILL.md",[],[],{"skillNo":95,"name":96,"nameZh":6,"description":97,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":19},"01M76TJVNAWDZ7DYQK3ZS0X09J","hermes-s6-container-supervision","Modify, debug, or extend the s6-overlay supervision tree inside the Hermes Agent Docker image — adding new services, debugging profile gateways, understanding the Architecture B main-program pattern.","https:\u002F\u002Fgithub.com\u002FSatangThevalue\u002Fai-skills\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fhermes-s6-container-supervision\u002FSKILL.md",[],[]]