cp-toolkit 2.2.10 → 2.2.11

package/README.md

@@ -124,6 +124,79 @@ applyTo: "**/*.ts,**/*.tsx"
 | `.github/instructions/*.instructions.md` | Path-specific rules |
 | `AGENTS.md` | Universal AI instructions |
 | `.vscode/mcp.json` | MCP server configuration |
+| `.github/cp-kit-models.yaml` | AI model allocation matrix |
+
+## Architect-Builder Strategy
+
+cp-toolkit implements the **Architect-Builder Pattern** for optimal AI agent performance. This strategy separates reasoning from execution:
+
+### Dual-Mode Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────┐
+│  SINGLE MODE                    HYBRID MODE (Architect-Builder) │
+│  ─────────────                  ───────────────────────────────  │
+│                                                                  │
+│  ┌─────────────────┐            PLANNER         EXECUTOR         │
+│  │ High-IQ Model   │            (Architect)     (Builder)        │
+│  │ Pure Reasoning  │            ┌──────────┐    ┌──────────┐     │
+│  └─────────────────┘            │ temp 0.1 │ →  │ temp 0.3 │     │
+│                                 │ Strategy │    │ Code Gen │     │
+│  • orchestrator                 └──────────┘    └──────────┘     │
+│  • security-auditor                                              │
+│  • debugger                     • backend-specialist             │
+│  • documentation-writer         • frontend-specialist            │
+│                                 • devops-engineer                │
+└─────────────────────────────────────────────────────────────────┘
+```
+
+### Mode Selection
+
+| Mode | Temperature | Use Case |
+|------|-------------|----------|
+| **Single** | 0.1 | Pure reasoning tasks (analysis, planning, auditing) |
+| **Hybrid Planner** | 0.1 | Strategic thinking, architecture decisions |
+| **Hybrid Executor** | 0.3 | Code generation, implementation |
+
+### Agent Categories
+
+1. **Leadership & Strategy** (Single Mode)
+   - `orchestrator`, `product-manager`, `product-owner`, `project-planner`
+   
+2. **Development Core** (Hybrid Mode)
+   - `backend-specialist`, `frontend-specialist`, `mobile-developer`, `game-developer`
+
+3. **Infrastructure & Ops** (Hybrid Mode)
+   - `devops-engineer`, `database-architect`, `security-auditor`, `penetration-tester`
+
+4. **Quality & Optimization** (Hybrid Mode)
+   - `qa-automation-engineer`, `test-engineer`, `performance-optimizer`, `debugger`
+
+5. **Specialists & Research** (Mixed)
+   - `code-archaeologist`, `documentation-writer`, `seo-specialist`, `explorer-agent`
+
+### Why This Works
+
+- **Planner (Architect)**: Uses high-reasoning models with low temperature for accurate strategic decisions
+- **Executor (Builder)**: Uses fast code-generation models with slightly higher temperature for creative implementation
+- **Cost Optimization**: Expensive models only for planning; economical models for bulk code generation
+- **Quality Assurance**: Clear separation prevents "hallucination drift" in long implementations
+
+### Models Configuration
+
+The `cp-kit-models.yaml` file defines the model allocation:
+
+```yaml
+agents:
+  backend-specialist:
+    mode: "hybrid"
+    planner:
+      model: "gpt-5.2"
+      task: "API architecture, security and data modeling"
+    executor:
+      model: "gpt-5.2-codex"
+      task: "Route and service implementation with perfect typing"
+```
 
 ## License
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cp-toolkit",
-  "version": "2.2.10",
+  "version": "2.2.11",
   "description": "Copilot Toolkit - AI Agent framework for GitHub Copilot, Claude, Gemini CLI, and other AI assistants",
   "keywords": [
     "ai-agents",