M.I.M.I.R - Multi-agent Intelligent Memory & Insight Repository

# Agent Chaining Architecture **Version**: 1.0.0 **Status**: Implemented (Phase 1 Complete) --- ## Usage ### Step 1: Generate Plan ```bash npm run chain "Draft migration plan for Docker" # Output: generated-agents/chain-output.md ``` ### Step 2: Execute Plan ```bash # Ensure MCP server is running npm run start:http # Terminal 1: http://localhost:3000/mcp # Execute tasks npm run execute generated-agents/chain-output.md # Terminal 2 ``` ## Flow ``` User Request ↓ PM (Analysis) ↓ Ecko (Optimize Tasks) ↓ PM (Assembly + Agent Roles) ↓ chain-output.md ↓ Agentinator (Generate Preambles) ↓ Execute Tasks Sequentially ``` ## Output Format Each task in the generated plan includes: - **Task ID**: Unique identifier for KG storage - **Agent Role**: Single-sentence agent background/specialization (e.g., "Backend engineer with Kafka experience, prefers simple PoC implementations") - **Recommended Model**: Best model for task type (GPT-4.1, Claude Sonnet 4, O3-mini, etc.) - **Optimized Prompt**: Complete, self-contained task instructions from Ecko - **Dependencies**: Task prerequisites - **Estimated Duration**: Time estimate **Agent roles feed into Agentinator** to generate specialized preambles for each task. --- ## Implementation **File**: `src/orchestrator/agent-chain.ts` **Agents**: 1. PM Agent (`claudette-pm.md`) - Task decomposition 2. Ecko Agent (`claudette-ecko.md`) - Prompt optimization 3. PM Agent - Final assembly + agent role definition --- ## Next: Knowledge Graph Integration Phase 2 will auto-store tasks in KG and enable worker execution.