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

# Mimir v2 Prompting Framework - Structure Summary **Created:** 2025-01-21 **Status:** ✅ Structure Complete - Ready for Content --- ## 📁 File Structure ``` docs/agents/v2/ ├── README.md # Framework overview & philosophy ├── STRUCTURE_SUMMARY.md # This file ├── 00-ecko-preamble.md # Stage 0: Prompt architect (optional) ├── 01-pm-preamble.md # Stage 1: Project manager ├── 02-agentinator-preamble.md # Stage 2: Dynamic preamble generator ├── 03-qc-failure-report-preamble.md # Stage 3: Failure analysis ├── 04-circuit-breaker-analysis-preamble.md # Stage 4: Runaway diagnosis ├── 05-final-report-preamble.md # Stage 5: Report synthesis └── templates/ ├── worker-template.md # Template for Worker agents (used by Agentinator) ├── qc-template.md # Template for QC agents (used by Agentinator) ├── reasoning-template.md # Standard reasoning format ├── verification-template.md # Standard verification format (TBD) └── output-format-template.md # Standard output format (TBD) ``` --- ## 🔄 Workflow & Agent Roles ### Static Preambles (Always the same) 1. **Ecko (00)** - OPTIONAL - Transforms raw user requests into structured prompts - Input: Vague user request - Output: Structured prompt for PM 2. **PM (01)** - REQUIRED - Decomposes requirements into task graph - Input: User requirements - Output: Task specifications with role descriptions 3. **Agentinator (02)** - REQUIRED (per task) - Generates custom Worker and QC preambles - Input: PM's role description + template - Output: Customized preamble for specific task 4. **QC Failure Report (03)** - AS NEEDED - Analyzes failures and provides corrections - Input: Failed task + QC report - Output: Actionable correction guidance 5. **Circuit Breaker Analysis (04)** - AS NEEDED - Diagnoses runaway executions - Input: Execution metadata - Output: Root cause + recommendations 6. **Final Report (05)** - REQUIRED - Synthesizes all results - Input: All task results - Output: Comprehensive report ### Dynamic Preambles (Generated per task) **Worker Preamble** - Generated by Agentinator - Template: `templates/worker-template.md` - Customized for each task based on PM's role description - Example: "Backend engineer with Node.js" → Node.js-specific worker preamble **QC Preamble** - Generated by Agentinator - Template: `templates/qc-template.md` - Customized for each task based on PM's QC role description - Example: "Senior QA with API testing" → API testing-specific QC preamble --- ## 🎯 Key Design Decisions ### Why Agentinator? **Problem:** PM would need to write full preambles for each task (1000+ tokens each) **Solution:** PM writes short role descriptions, Agentinator expands them into full preambles **Benefits:** - ✅ PM output stays concise (50 tokens vs. 1000+ tokens per task) - ✅ Consistency (all agents use same template structure) - ✅ Customization (each agent gets task-specific guidance) - ✅ Quality (templates encode best practices from research) - ✅ Maintainability (update template once, affects all future agents) ### Why Templates? **Problem:** Every worker/QC agent needs same structure but different content **Solution:** Templates with `<TO BE DEFINED>` placeholders that Agentinator fills **Benefits:** - ✅ Enforces consistent structure (all 8 sections present) - ✅ Encodes best practices (reasoning, verification, bookending) - ✅ Easy to improve (update template, all future agents benefit) - ✅ Reduces errors (structure is guaranteed correct) --- ## 📊 Token Economics ### Old Approach (v1): ``` PM Output per task: ~1500 tokens (includes full preamble) × 5 tasks = 7,500 tokens ``` ### New Approach (v2): ``` PM Output per task: ~200 tokens (role description only) × 5 tasks = 1,000 tokens Agentinator per task: ~500 tokens (template expansion) × 5 tasks = 2,500 tokens Total: 3,500 tokens (53% reduction!) ``` --- ## 🔧 Implementation Status ### ✅ Complete: - [x] Framework README - [x] Ecko preamble boilerplate (00) - [x] PM preamble boilerplate (01) - [x] Agentinator preamble boilerplate (02) - [x] Worker template boilerplate - [x] QC Failure Report preamble boilerplate (03) - [x] Circuit Breaker Analysis preamble boilerplate (04) - [x] Final Report preamble boilerplate (05) - [x] Reasoning template ### ⏳ To Do: - [ ] QC template boilerplate (in templates/) - [ ] Verification template - [ ] Output format template - [ ] Fill in `<TO BE DEFINED>` sections for each preamble - [ ] Add examples to each preamble - [ ] Test with actual execution - [ ] Measure success rate improvement --- ## 🎯 Next Steps 1. **Create QC Template** - Mirror worker template structure 2. **Fill PM Preamble** - Priority 1 (most critical) 3. **Fill Worker Template** - Priority 2 (used by all tasks) 4. **Fill QC Template** - Priority 3 (used by all tasks) 5. **Fill Agentinator** - Priority 4 (orchestrates template usage) 6. **Test Execution** - Run with v2 prompts 7. **Measure & Iterate** - Track success rate, refine --- **Status:** 🚧 Structure complete, ready for content **Next:** Fill in PM preamble (Stage 1)