UNPKG

automagik-genie

Version:

Self-evolving AI agent orchestration framework with Model Context Protocol support

166 lines (136 loc) 5.14 kB
# Create Collective - Shape-Shifting Human-World Intelligence ## Philosophy Create is **not** a fixed library of pre-built agents and spells. Create is a **generative intelligence** that becomes what the user needs, when they need it. ## Core Architecture ### Minimal Core (Always Present) - **researcher** - Information gathering, source validation - **writer** - Content creation (any format, any domain) - **editor** - Content refinement and polish - **install** - Setup and initialization ### Generative Capabilities (Core Meta-Spells) - **shape-shifting.md** - Fluid adaptation to any human-world task - **agent-generation.md** - Generate new agents from usage patterns - **spell-generation.md** - Create domain spells on-demand - **workflow-generation.md** - Build repeatable processes organically ### Everything Else = Generated On-Demand ## How It Works ### Example 1: Project Manager Emerges ``` User requests sprint planning help 3 times ↓ Create recognizes pattern ↓ Generates project-manager agent with sprint-planning spell ↓ Agent now available for all future sprint work ↓ Continues evolving with each use ``` ### Example 2: Crisis Communication ``` User: "We had a security breach, need crisis comms" ↓ Create doesn't have crisis-communication spell ↓ Generates spell in real-time from best practices ↓ Applies immediately to user's crisis ↓ Spell now part of Create's knowledge for future crises ``` ### Example 3: Domain Expertise ``` User regularly works with financial models ↓ Create generates finance-analyst capabilities ↓ Becomes expert in user's financial domain ↓ Other users in same domain benefit from generated expertise ``` ## Generative Intelligence Process 1. **Pattern Recognition** - Identify user needs (first-time or recurring) 2. **Capability Generation** - Create agent/spell/workflow on-demand 3. **Immediate Application** - Solve current problem 4. **Learning & Refinement** - Improve with each use 5. **Maturation** - Experimental → Validated → Core ## Coverage Domains Create can shape-shift into: - ✅ Executive assistant (calendar, email, tasks) - ✅ Project manager (sprints, roadmaps, status) - ✅ Business writer (proposals, decks, sales copy) - ✅ Strategist (analysis, planning, frameworks) - ✅ Communicator (internal/external, crisis, PR) - ✅ Analyst (data, reports, insights) - ✅ HR specialist (JDs, reviews, onboarding) -**Any human-world role the user needs** ## Why This Approach? ### Traditional Approach (❌ Rejected) ``` Pre-build 50+ agents, 100+ spells, 50+ workflows Result: Bloated, rigid, many capabilities unused Maintenance: Constant upkeep of unused features ``` ### Generative Approach (✅ Create's Way) ``` Start with 4 core agents + 4 meta-spells Generate capabilities from actual usage Result: Lean, fluid, perfectly matched to user needs Maintenance: Only maintain what's actually used ``` ## File Structure ``` .genie/create/ ├── AGENTS.md (orchestrator identity - shape-shifting core) ├── README.md (this file - philosophy and architecture) ├── routing.md (delegation guidance) ├── qa.md (validation standards) │ ├── spells/ │ ├── shape-shifting.md ⭐ CORE │ ├── agent-generation.md ⭐ CORE │ ├── spell-generation.md ⭐ CORE │ ├── workflow-generation.md ⭐ CORE │ ├── prompting-standards-create.md │ ├── content-evidence.md │ ├── style-guide-integration.md │ ├── asset-naming-rules.md │ └── publishing-workflow.md │ ├── agents/ │ ├── researcher.md ⭐ CORE │ ├── writer.md ⭐ CORE │ ├── editor.md ⭐ CORE │ ├── install.md ⭐ CORE │ ├── wish.md │ └── [generated agents emerge here...] │ ├── workflows/ │ ├── wish.md │ ├── forge.md │ ├── install.md │ └── [generated workflows emerge here...] │ └── teams/ └── creative-council/ (advisory, not execution) ``` ## Success Metrics - **Fluidity:** How quickly Create adapts to new domains - **Coverage:** % of user requests Create can handle (goal: >90%) - **Learning:** How many capabilities generated and matured - **User Satisfaction:** Does Create feel like the expert they need? ## Comparison to Code Collective | Aspect | Code Collective | Create Collective | |--------|-----------------|-------------------| | Domain | Software engineering | All human-world work | | Capabilities | Pre-built (git, tests, refactor, etc.) | Generated on-demand | | Approach | Specialized from start | Generalist → specialist | | Evolution | Mostly static | Continuously evolving | | Scope | Narrow, deep | Broad, adaptive | **Together:** Code handles all software work, Create handles all human work. **Result:** Complete Genie intelligence for any professional task. ## The Vision Create is not an assistant with fixed capabilities. **Create is an intelligence that becomes the expert you need, when you need it.** This is shape-shifting mastery applied to human-world work.