{
  "markdown": "# Agentic Design Patterns — Gemini CLI Extension\n\nA Gemini CLI **Extension** packaging 28 agentic design pattern skills. Install with a single command and start building production-ready AI agents immediately.\n\n## Install\n\n```bash\ngemini extensions install https://github.com/hajekim/agentic-design-patterns-extension\n```\n\nAfter installation, restart Gemini CLI. The 28 skills activate automatically when you describe what you want to build.\n\n> Current version: **v2.2.5** — See [CHANGELOG.md](CHANGELOG.md) for full version history.\n\n## What This Extension Provides\n\n**28 agent skills** across four categories:\n\n| Category | Skills |\n|----------|--------|\n| Core Patterns | Prompt Chaining, Routing, Parallelization, Reflection, Tool Use, Planning, Multi-Agent Collaboration |\n| State Management | Memory Management, Learning & Adaptation, MCP, Goal Setting |\n| Reliability | Exception Handling, Human-in-the-Loop, RAG |\n| Advanced Patterns | A2A, Resource-Aware, Reasoning, Guardrails, Evaluation, Prioritization, Exploration |\n| Appendix | Prompt Engineering, GUI Agents, Agentic Frameworks, AgentSpace, AI CLI, Coding Agents, Reasoning Engines |\n\nSkills use the **DEFINE → PLAN → ACTION** workflow and include implementation examples in Google ADK, LangChain, and LangGraph.\n\n## How Skills Activate\n\nGemini CLI reads the `name` and `description` of each skill. When your request matches a skill's description, the model automatically loads the full skill instructions.\n\n**1,376 trigger phrases** across four languages:\n\n```\n# English\n\"Build a multi-step agent pipeline\"          → Prompt Chaining\n\"set up MCP server\"                          → MCP\n\"choose agent framework\"                     → Agentic Frameworks\n\"thinking model for complex reasoning\"       → Reasoning Engines\n\n# 한국어\n\"프롬프트 체이닝으로 파이프라인 만들어줘\"    → Prompt Chaining\n\"MCP 구성을 해줘\"                           → MCP\n\"메모리뱅크 만들어줘\"                       → Memory Management\n\"추론 모델 언제 써야 해?\"                   → Reasoning Engines\n\n# 日本語\n\"マルチエージェントを構築したい\"             → Multi-Agent Collaboration\n\"MCPサーバーを設定したい\"                   → MCP\n\"RAGパイプラインを作りたい\"                 → RAG\n\n# 中文\n\"帮我构建多智能体系统\"                      → Multi-Agent Collaboration\n\"配置MCP服务器\"                            → MCP\n\"搭建RAG知识库问答系统\"                    → RAG\n```\n\n## Commands & Agents\n\n### Slash Commands\n\n```bash\n# Browse all 28 patterns grouped by category\n/pattern-summary\n\n# Filter by category: core / state / reliability / advanced / appendix\n/pattern-summary reliability\n\n# Look up a specific pattern\n/pattern-summary planning\n\n# Generate a Python code skeleton for a pattern\n/gen-skeleton planning\n/gen-skeleton rag\n```\n\n**`/gen-skeleton planning` example output:**\n\n```python\nfrom google import genai\nfrom google.adk.agents import LlmAgent\nfrom google.adk.runners import Runner\nfrom google.adk.sessions import InMemorySessionService\nimport asyncio\n\nplanner = LlmAgent(\n    name=\"planner\",\n    model=\"gemini-2.5-flash\",\n    instruction=\"\"\"You are a planning agent. Given a complex goal, decompose it\n    into an ordered list of concrete subtasks. For each subtask specify:\n    what to do, what input it needs, and what output it produces.\"\"\",\n)\n\nasync def run(goal: str) -> str:\n    session_service = InMemorySessionService()\n    await session_service.create_session(\n        app_name=\"planning-demo\", user_id=\"user\", session_id=\"s1\"\n    )\n    runner = Runner(\n        agent=planner, app_name=\"planning-demo\", session_service=session_service\n    )\n    async for event in runner.run_async(\n        user_id=\"user\", session_id=\"s1\",\n        new_message=genai.types.Content(\n            role=\"user\", parts=[genai.types.Part(text=goal)]\n        ),\n    ):\n        if event.is_final_response():\n            return event.content.parts[0].text\n    return \"\"\n\nif __name__ == \"__main__\":\n    print(asyncio.run(run(\"Build a customer support bot for returns, billing, and tech issues\")))\n```\n\n### Sub-agents (Preview)\n\n**`@architect`** — recommends an optimal pattern combination for your problem.\n\n```\nInput:  natural-language problem description\n\nOutput:\n  - Problem analysis (scope, constraints)\n  - Recommended patterns with rationale\n  - How the patterns combine in a system design\n  - Next step: /gen-skeleton <primary-pattern>\n```\n\n```bash\n@architect \"I need to build a customer support bot that learns from feedback\"\n```\n\n**`@reviewer`** — reviews agent code for pattern compliance and SDK conventions.\n\n```\nInput:  Python agent code (paste directly)\n\nOutput:\n  - Pattern compliance checklist (12 core/state/reliability patterns)\n  - SDK convention check (LlmAgent, Runner, InMemorySessionService, google-genai)\n  - Issues found with file/line references\n  - Concrete fix recommendations\n```\n\n```bash\n@reviewer\n# then paste your code\n```\n\n**Recommended workflow:**\n1. `@architect` → get pattern recommendations\n2. `/gen-skeleton <pattern>` → generate code skeleton\n3. `@reviewer` → verify implementation\n\n---\n\n## Extension Management\n\n```bash\n# Check installed extensions and their skills\ngemini extensions list\n\n# Update to latest version\ngemini extensions update agentic-design-patterns\n\n# Disable without uninstalling\ngemini extensions disable agentic-design-patterns\n\n# Uninstall\ngemini extensions uninstall agentic-design-patterns\n```\n\n## Platform Compatibility\n\n| Platform | Installation | Activation |\n|----------|-------------|------------|\n| **Gemini CLI** | `gemini extensions install <url>` | Semantic — model reads description autonomously |\n| **Antigravity** | Copy `skills/` to `.agents/skills/` | Keyword pattern matching |\n| **Claude Code** | Symlink `skills/` to `.claude/skills/` | Semantic judgment + slash commands |\n\n> For Antigravity and Claude Code, use the [Skills-only version](https://github.com/hajekim/agentic-design-patterns-skills).\n\n## Extension Structure\n\n```\nagentic-design-patterns/\n├── gemini-extension.json     ← Extension manifest (v2.2.4)\n├── GEMINI.md                 ← Global context: pattern guide, model guide, tech decisions\n├── mcp_server.py             ← Skill-search MCP server (list_patterns, get_skill, search_skills)\n├── commands/\n│   ├── gen-skeleton.toml    ← /gen-skeleton <pattern> — generate code skeleton\n│   └── pattern-summary.toml ← /pattern-summary [filter] — browse patterns\n├── agents/\n│   ├── architect.md         ← Recommend optimal pattern combinations\n│   └── reviewer.md          ← Review code for pattern compliance\n└── skills/                   ← 28 skill definitions\n    ├── planning/\n    │   └── SKILL.md\n    ├── rag/\n    │   └── SKILL.md\n    └── ...\n```\n\n## Trigger Coverage\n\n| Language | Count |\n|----------|------:|\n| English  | 474   |\n| Korean   | 337   |\n| Japanese | 284   |\n| Chinese  | 281   |\n| **Total**| **1,376** |\n\n## Model Reference\n\n| Task Type | Recommended Model | Thinking Budget |\n|-----------|------------------|----------------|\n| Simple pipelines — prompt-chaining, routing | `gemini-2.5-flash-lite` | Not supported |\n| Medium complexity — tool-use, RAG, parallelization | `gemini-2.5-flash` | Dynamic (leave unset) |\n| Complex reasoning — planning, reasoning, evaluation | `gemini-2.5-flash` or `gemini-2.5-pro` | Set high |\n| Large-scale coordination — multi-agent, a2a | `gemini-2.5-pro` | Set high |\n\nThinking Budget is an adjustable reasoning depth parameter available on Flash and Pro models (not Flash-Lite). Leave it unset for most tasks — the model decides dynamically.\n\n## Source\n\nBased on **\"Agentic Design Patterns\"** by Antonio Gulli (424 pages, 21 chapters + 6 appendices).\n\n## License\n\nMIT License — free to use, modify, and distribute.\n",
  "bytes": 7574,
  "sha": "3eb650fc63da2b5ee61564a4ecdf33a67dc0bc1ac26000ec10cd53096a6604ca",
  "repo_slug": "hajekim/agentic-design-patterns-extension",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_hajekim_agentic_design_patterns_extensio_1e0203f8/readme"
}