feat: simplify centralized rules with sync-based approach

Replace complex gateway pattern with simple Git-based sync strategy that:
- Auto-generates CLAUDE.md, AGENTS.md, .cursorrules from central repository
- Auto-detects language (Python, TypeScript, Go, etc.) and framework (React, Django, etc.)
- Works offline after initial sync (no API calls required)
- Compatible with all AI assistants (Claude Code, Cursor, Gemini)
- Consolidates documentation into single comprehensive README

Changes:
- New sync-strategy/ with sync-ai-rules.sh (auto-detection & generation)
- Quick-start helper: create-central-repo.sh
- Example rules: Python, React, universal security/testing standards
- Moved old gateway approach to gateway-strategy/ (advanced use cases)
- Updated experiments/README.md pattern description
- Consolidated centralized-rules/README.md (removed duplicate docs)

Benefits over old approach:
- 100 lines bash vs 500+ lines TypeScript
- No Node.js service to deploy
- Works with standard AI config files
- Language/framework-aware rule selection

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Author: PaulDuvall

experiments/examples/centralized-rules/README.md

@@ -0,0 +1,122 @@
+# Centralized Rules Example
+
+This example demonstrates the three-layer architecture for centralizing AI rules across an organization using the Claude SDK.
+
+## Architecture
+
+```
+┌─────────────┐     ┌─────────────┐     ┌─────────────┐
+│  ai-dev-cli │ ──► │  ai-gateway │ ──► │  Claude API │
+└─────────────┘     └─────────────┘     └─────────────┘
+       │                   │
+       │            ┌──────┴──────┐
+       │            │  Org Rules  │
+       │            │  Logging    │
+       │            │  Filters    │
+       │            └─────────────┘
+       │
+       └──────────► @yourorg/org-ai-client (alternative)
+```
+
+## Components
+
+### 1. ai-gateway/
+
+Internal service that owns all org rules and calls the Claude API.
+
+**Features**:
+- Centralized system prompts with org rules
+- Request/response logging
+- Input/output filtering hooks
+
+### 2. org-ai-client/
+
+Shared SDK wrapper that embeds org rules. Use this if you don't want a network service yet.
+
+**Features**:
+- Org rules baked into system prompts
+- Consistent interface across all repos
+- Version-controlled rule updates
+
+### 3. ai-dev-cli/
+
+Developer CLI that calls the gateway or wrapper library.
+
+**Features**:
+- Simple commands: `ai-dev plan`, `ai-dev refactor`
+- Auto-detects repo context
+- Consistent UX across tools
+
+## Setup
+
+### Prerequisites
+
+- Node.js 18+
+- `ANTHROPIC_API_KEY` environment variable
+
+### Install dependencies
+
+```bash
+cd ai-gateway && npm install
+cd ../org-ai-client && npm install
+cd ../ai-dev-cli && npm install
+```
+
+### Run the gateway
+
+```bash
+cd ai-gateway
+npm start
+# Gateway running on http://localhost:3000
+```
+
+### Use the CLI
+
+```bash
+cd ai-dev-cli
+npm link
+ai-dev plan "Implement idempotent refund API"
+```
+
+## Customization
+
+### Modify org rules
+
+Edit the system prompt in `ai-gateway/src/claudeClient.ts` or `org-ai-client/src/index.ts`:
+
+```typescript
+const systemPrompt = `
+You are the org AI pair programmer.
+Follow ORG RULES v1.4:
+- Use spec-first development.
+- Your custom rules here...
+`;
+```
+
+### Add input/output filters
+
+Extend the gateway with validation:
+
+```typescript
+// Before calling Claude
+if (containsSecrets(input)) {
+  throw new Error("Input contains secrets");
+}
+
+// After receiving response
+if (containsBannedPatterns(response)) {
+  throw new Error("Response violates policy");
+}
+```
+
+### Integrate policy-as-code
+
+Add OPA or Cedar policy checks:
+
+```typescript
+const allowed = await opa.evaluate("ai/task_allowed", {
+  repo: input.repo,
+  taskType: input.taskType,
+  user: req.user
+});
+```