Agent Definitions for Future Development\n\n## What are Agents?\n\nAgents are specialized sub-AI assistants that handle specific complex tasks autonomously. When you ask for a task that matches an agent's expertise, use the agent to handle it.\n\n## Available Agents\n\n### 1. llm-integration Agent\n\nPurpose: Add new LLM providers and multi-model support\n\nWhen to use:\n- \"Add OpenAI provider\"\n- \"Switch to Claude for better quality\"\n- \"Implement load balancing across providers\"\n- \"Add support for vision APIs\"\n\nWhat it does:\n- Creates provider implementation\n- Sets up configuration\n- Writes tests\n- Updates documentation\n\nInvocation:\n\nRunSubagent(agentName=\"llm-integration\", \nprompt=\"Add OpenAI provider with cost tracking\")\n\n\nUses: Skill llm-integration, Prompt add-provider.md\n\n---\n\n### 2. performance-tuning Agent\n\nPurpose: Optimize latency and cost through caching, batching, and optimization\n\nWhen to use:\n- \"My latency is 1 second, reduce it\"\n- \"AI costs are $500/month, find optimizations\"\n- \"Implement caching for product search\"\n- \"Set up Redis for distributed caching\"\n\nWhat it does:\n- Analyzes current performance\n- Identifies bottlenecks\n- Implements optimizations\n- Measures impact\n\nInvocation:\n\nRunSubagent(agentName=\"performance-tuning\",\nprompt=\"Current latency: 800ms. Reduce to <500ms\")\n\n\nUses: Skill performance-optimization, Prompt optimize-performance.md\n\n---\n\n### 3. mkdocs-content Agent\n\nPurpose: Create and update MkDocs documentation\n\nWhen to use:\n- \"Document this new feature\"\n- \"Create architecture diagram\"\n- \"Update the decision matrix\"\n- \"Reorganize docs structure\"\n\nWhat it does:\n- Creates/updates markdown files\n- Maintains consistency\n- Updates navigation (mkdocs.yml)\n- Adds mermaid diagrams\n\nInvocation:\n\nRunSubagent(agentName=\"mkdocs-content\",\nprompt=\"Document new review sentiment use case with decision matrix\")\n\n\nUses: Skill documentation, MkDocs conventions\n\n---\n\n### 4. test-generation Agent\n\nPurpose: Generate comprehensive unit and integration tests\n\nWhen to use:\n- \"Write tests for new service\"\n- \"Create integration test suite\"\n- \"Improve test coverage to 85%\"\n- \"Add error handling tests\"\n\nWhat it does:\n- Analyzes code\n- Creates test cases\n- Follows existing patterns\n- Implements mocking\n\nInvocation:\n\nRunSubagent(agentName=\"test-generation\",\nprompt=\"Create tests for ReviewSentimentService\")\n\n\nUses: Skill testing-strategy\n\n---\n\n### 5. Explore Agent\n\nPurpose: Fast codebase exploration and Q&A\n\nWhen to use:\n- \"What endpoints do we have?\"\n- \"How does caching work in product search?\"\n- \"Which files handle error handling?\"\n- \"What DTOs are used in recommendations?\"\n\nWhat it does:\n- Searches codebase\n- Reads files\n- Answers specific questions\n- Safe (read-only)\n\nInvocation:\n\nRunSubagent(agentName=\"Explore\",\nprompt=\"Find all AI service implementations and summarize their patterns\")\n\n\nThoroughness: quick, medium, or thorough\n\n---\n\n## How to Use Agents\n\n### In Copilot Chat\n\n@agent-name\nDescribe your task\n\n\n### Programmatically\npython\nfrom copilot_sdk import run_subagent\n\nresult = run_subagent(\n agent_name=\"llm-integration\",\n prompt=\"Add Anthropic Claude 3 provider with cost tracking\",\n context={\"project_path\": \"/path/to/project\"}\n)\n\n\n## Agent Decision Tree\n\n\nWhat do you need?\n├─ Add new LLM provider → llm-integration agent\n├─ Reduce latency/cost → performance-tuning agent \n├─ Create/update docs → mkdocs-content agent\n├─ Write tests → test-generation agent\n├─ Explore codebase → Explore agent\n└─ Something else → Ask Copilot directly\n\n\n## Best Practices\n\n### ✅ Do\n- Be specific in your prompt\n- Give context (current state, goals)\n- Let agent work autonomously\n- Review agent's work before committing\n- Provide feedback if needed\n\n### ❌ Don't\n- Interrupt agent mid-task\n- Give vague prompts\n- Use wrong agent for task\n- Skip code review\n- Make conflicting requests\n\n---\n\n## Example: Using Multiple Agents\n\nScenario: Add new feature end-to-end\n\n1. Use Explore agent\n \n \"Show me similar use cases to base this on\"\n\n\n2. Use llm-integration agent (if needed)\n \n \"Add support for Anthropic Claude for this use case\"\n\n\n3. Code manually or ask Copilot\n \n \"Create ReviewSentimentService following the pattern\"\n\n\n4. Use test-generation agent\n \n \"Create comprehensive tests for ReviewSentimentService\"\n\n\n5. Use performance-tuning agent (if needed)\n \n \"Optimize the latency, current P95 is 800ms\"\n\n\n6. Use mkdocs-content agent\n \n \"Document this new feature with architecture diagram\"\n\n\n---\n\n## Agent Capabilities Summary\n\n| Agent | Code? | Docs? | Tests? | Optimize? | Search? |\n|-------|-------|-------|--------|-----------|----------|\n| llm-integration | ✅ | ✅ | ✅ | ❌ | ❌ |\n| performance-tuning | ✅ | ✅ | ✅ | ✅ | ❌ |\n| mkdocs-content | ❌ | ✅ | ❌ | ❌ | ⚠️ |\n| test-generation | ✅ | ⚠️ | ✅ | ❌ | ❌ |\n| Explore | ❌ | ❌ | ❌ | ❌ | ✅ |\n\n---\n"