{
  "markdown": "# 🐝 gem-swarm\n\nMulti-agent orchestration extension for [Gemini CLI](https://github.com/google-gemini/gemini-cli).\n\n20 domain-specialist agents · 43 skills · 13 commands · parallel subagent dispatch · persistent sessions\n\n---\n\n## Quick Start\n\n```bash\n# 1. Enable agents in Gemini CLI\n# Add to ~/.gemini/settings.json:\n# { \"experimental\": { \"enableAgents\": true } }\n\n# 2. Install\ngemini extension install github:<your-username>/gem-swarm\n\n# 3. Verify\ngemini\n> /gem-swarm:status\n```\n\n**That's it.** You now have 20 specialist agents ready to work.\n\n### Быстрый старт (RU)\n\n```bash\n# 1. Включи агентов в настройках Gemini CLI\n# Добавь в ~/.gemini/settings.json:\n# { \"experimental\": { \"enableAgents\": true } }\n\n# 2. Установи\ngemini extension install github:<your-username>/gem-swarm\n\n# 3. Проверь\ngemini\n> /gem-swarm:status\n```\n\n**Готово.** 20 агентов-специалистов к вашим услугам.\n\n---\n\n## How to Use / Как пользоваться\n\ngem-swarm gives you **three ways** to work, from fastest to most thorough:\n\ngem-swarm даёт **три способа** работы — от самого быстрого до самого детального:\n\n### 1. `@agent` — Direct Call / Прямой вызов ⚡\n\n**The fastest way.** Call a specialist directly — no orchestrator, no questions, no phases. The agent just does the work.\n\n**Самый быстрый способ.** Вызываешь специалиста напрямую — без оркестратора, без вопросов, без фаз. Агент просто делает.\n\n```\n@frontend_specialist Build a signup form with email validation\n@debugger Fix the crash in auth.ts — stack trace attached\n@test_engineer Write unit tests for UserService\n@security_auditor Audit the /api/payments endpoint\n@backend_specialist Add rate limiting to the API\n```\n\n**When to use / Когда использовать:**\n- You know exactly what you need / Ты точно знаешь что нужно\n- Single-domain task / Задача в одной области\n- You want results fast, not a plan / Хочешь результат, а не план\n\n### 2. `/gem-swarm:command` — Structured Commands / Структурированные команды\n\nPredefined workflows for common tasks. More structured than `@agent`, but no full orchestration overhead.\n\nГотовые сценарии для типовых задач. Структурированнее чем `@agent`, но без нагрузки полной оркестрации.\n\n| Command | What it does |\n|---------|-------------|\n| `/gem-swarm:debug` | Systematic debugging with hypothesis-driven investigation |\n| `/gem-swarm:review` | Deep code review — logic, security, structure, performance |\n| `/gem-swarm:test` | Generate and run tests |\n| `/gem-swarm:security-audit` | OWASP 2025, supply chain, attack surface analysis |\n| `/gem-swarm:deploy` | Production deployment with pre-flight checks |\n| `/gem-swarm:create` | Create a new application from natural language |\n| `/gem-swarm:enhance` | Add or update features in an existing project |\n| `/gem-swarm:plan` | Project planning — analysis, breakdown, dependencies |\n| `/gem-swarm:status` | View active session status |\n| `/gem-swarm:resume` | Resume an interrupted session |\n| `/gem-swarm:archive` | Archive the active session |\n\n### 3. `/gem-swarm:orchestrate` — Full Orchestration / Полная оркестрация 🎯\n\nThe **full 4-phase workflow** with design dialogue, implementation planning, multi-agent execution, and code review. Use for complex, multi-domain tasks.\n\n**Полный 4-фазный воркфлоу**: диалог по дизайну, планирование, мульти-агентное выполнение, код-ревью. Для сложных задач.\n\n```\n/gem-swarm:orchestrate Build a dashboard with auth, real-time data, and responsive design\n```\n\n**When to use / Когда использовать:**\n- Complex task spanning multiple domains / Сложная задача из нескольких областей\n- You need a plan before execution / Нужен план перед выполнением\n- Multiple agents need to coordinate / Нескольким агентам нужно координировать работу\n\n---\n\n## When to Use What / Когда что использовать\n\n| Situation | Best choice | Why |\n|-----------|------------|-----|\n| \"Add a button\" | `@frontend_specialist` | Simple, one agent, fast |\n| \"Find this bug\" | `@debugger` or `/gem-swarm:debug` | Focused investigation |\n| \"Review my PR\" | `/gem-swarm:review` | Structured review workflow |\n| \"Build a new feature\" | `/gem-swarm:orchestrate` | Multi-agent, needs planning |\n| \"New project from scratch\" | `/gem-swarm:orchestrate` | Design → Plan → Build → Review |\n| \"Is this code secure?\" | `@security_auditor` | Quick audit, one domain |\n| \"Full security audit\" | `/gem-swarm:security-audit` | Thorough, structured audit |\n\n**Rule of thumb / Правило**: If you can describe the task in one sentence for one specialist → use `@agent`. If it needs coordination → use `/gem-swarm:orchestrate`.\n\n**Правило**: Если задачу можно описать одним предложением для одного специалиста → `@agent`. Если нужна координация → `/gem-swarm:orchestrate`.\n\n---\n\n## Agent Roster / Список агентов\n\n| Agent | Domain | Example call |\n|-------|--------|-------------|\n| `frontend_specialist` | React/Next.js, UI/UX, responsive design | `@frontend_specialist` |\n| `backend_specialist` | API, server architecture, backend logic | `@backend_specialist` |\n| `database_architect` | Schema design, queries, migrations, indexing | `@database_architect` |\n| `security_auditor` | OWASP 2025, supply chain, zero trust | `@security_auditor` |\n| `penetration_tester` | Offensive security, exploit analysis | `@penetration_tester` |\n| `test_engineer` | Unit/integration testing, TDD, coverage | `@test_engineer` |\n| `qa_automation_engineer` | E2E testing, Playwright, CI pipelines | `@qa_automation_engineer` |\n| `debugger` | Root cause analysis, crash investigation | `@debugger` |\n| `performance_optimizer` | Profiling, Core Web Vitals, bundle size | `@performance_optimizer` |\n| `devops_engineer` | CI/CD, deployment, server management | `@devops_engineer` |\n| `mobile_developer` | React Native, Flutter, mobile patterns | `@mobile_developer` |\n| `game_developer` | Game mechanics, engines (Unity/Godot/Phaser) | `@game_developer` |\n| `documentation_writer` | Technical docs, READMEs, API docs | `@documentation_writer` |\n| `product_manager` | Requirements, user stories, prioritization | `@product_manager` |\n| `product_owner` | Roadmap, backlog, stakeholder management | `@product_owner` |\n| `project_planner` | Task breakdown, dependency graphs | `@project_planner` |\n| `explorer_agent` | Codebase discovery, architectural analysis | `@explorer_agent` |\n| `code_archaeologist` | Legacy code analysis, codebase archaeology | `@code_archaeologist` |\n| `seo_specialist` | SEO, GEO, E-E-A-T, Core Web Vitals | `@seo_specialist` |\n| `website_cloner` | Pixel-perfect website cloning, 5-phase workflow | `@website_cloner` |\n\n---\n\n## Architecture\n\n```\nUser\n ├── @agent ──────────────── Agent (direct, fast)\n ├── /gem-swarm:command ──── Structured workflow → Agent(s)\n └── /gem-swarm:orchestrate → Orchestrator (GEMINI.md)\n                                    │\n                                    ├── MCP Server (maestro-server.js)\n                                    │   ├── Workspace tools (init, validate, settings)\n                                    │   ├── Session tools (create, status, transition, archive)\n                                    │   └── Content tools (skills, agents, runtime context)\n                                    │\n                                    ├── Subagent dispatch (native Gemini CLI agents)\n                                    │   └── 20 specialists with per-agent tools, temperature, turn limits\n                                    │\n                                    ├── Hooks (SessionStart, BeforeAgent, AfterAgent, SessionEnd)\n                                    └── Policies (shell command safety rules)\n```\n\n### 4-Phase Workflow (orchestrate mode)\n\n1. **Design** — Interactive design dialogue with approval gates\n2. **Planning** — Implementation plan with dependency analysis and validation\n3. **Execution** — Parallel or sequential agent dispatch\n4. **Completion** — Code review, session archival, summary\n\n---\n\n## Configuration\n\nSettings via environment variables or extension `.env` file:\n\n| Setting | Env Var | Default | Description |\n|---------|---------|---------|-------------|\n| Disabled Agents | `GEM_SWARM_DISABLED_AGENTS` | — | Comma-separated agents to exclude |\n| Max Retries | `GEM_SWARM_MAX_RETRIES` | `2` | Phase retry limit |\n| Auto Archive | `GEM_SWARM_AUTO_ARCHIVE` | `true` | Archive on success |\n| Validation | `GEM_SWARM_VALIDATION_STRICTNESS` | `normal` | Validation gating mode |\n| State Directory | `GEM_SWARM_STATE_DIR` | `docs/gem-swarm` | Session and plan state root |\n| Max Concurrent | `GEM_SWARM_MAX_CONCURRENT` | `0` | Parallel batch size (0 = all) |\n| Execution Mode | `GEM_SWARM_EXECUTION_MODE` | `ask` | `parallel`, `sequential`, or `ask` |\n\n---\n\n## Development\n\n```bash\n# Link for local development\ngit clone https://github.com/<your-username>/gem-swarm.git\ncd gem-swarm\ngemini extension install .\n\n# Regenerate agent stubs and registries\nnpm run generate\n\n# Verify MCP server loads\nnode -e \"require('./src/mcp/maestro-server')\"\n```\n\n## Technical Details\n\n- **Runtime:** Node.js ≥ 20 (no external dependencies)\n- **Protocol:** MCP JSON-RPC 2.0 over STDIO\n- **State:** Markdown-based persistent sessions\n- **Security:** TOML policy rules for shell command safety\n- **License:** Apache 2.0 — see [LICENSE](LICENSE)\n",
  "bytes": 9189,
  "sha": "5c49bd51941de3d836c21ff44c5d0a556e01cb4c6891d78d9815814e1f2e1a42",
  "repo_slug": "yqecea/gem-swarm",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_yqecea_gem_swarm_d20c8e35/readme"
}