feat(llm): add LLM call support to worker IPC chain

- ProviderManager: align API with contracts ProviderAdapter
- WorkerProtocol: add llm.request/llm.response message types
- WorkerRuntime: add call_llm() for worker→parent→LLM flow
- WorkerManager: support tool_registry and provider_manager injection

P1-1 complete, P1-2 protocol layer complete.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

packages/workers/src/WorkerRuntime.ts

@@ -24,6 +24,20 @@ export interface ToolCallResult {
   content: Record<string, unknown>
 }
 
+export interface LLMRequest {
+  messages: Array<{ role: string; content: unknown }>
+  model?: string
+  max_tokens?: number
+  temperature?: number
+  tools?: unknown[]
+}
+
+export interface LLMResponse {
+  content: string
+  usage?: { input_tokens: number; output_tokens: number }
+  tool_calls?: Array<{ name: string; arguments: Record<string, unknown> }>
+}
+
 export class WorkerRuntime {
   private agent_id: string
   private session_id: string
@@ -63,6 +77,36 @@ export class WorkerRuntime {
     return promise
   }
 
+  /**
+   * Call LLM through parent process via IPC.
+   * INV-3: This is the ONLY way workers access LLM.
+   */
+  async call_llm(request: LLMRequest): Promise<LLMResponse> {
+    const call_id = crypto.randomUUID()
+
+    const promise = new Promise<LLMResponse>((resolve, reject) => {
+      this.pending_calls.set(call_id, { resolve: resolve as any, reject })
+
+      // Set timeout (LLM calls can be long)
+      setTimeout(() => {
+        this.pending_calls.delete(call_id)
+        reject(new Error(`LLM call timeout: ${request.model || 'default'}`))
+      }, 300000) // 5 minutes
+    })
+
+    // Send llm.request via IPC
+    this.send_message('llm.request', {
+      call_id,
+      messages: request.messages,
+      model: request.model || 'claude-haiku-4-5-20251001',
+      max_tokens: request.max_tokens || 4096,
+      temperature: request.temperature,
+      tools: request.tools
+    })
+
+    return promise
+  }
+
   /**
    * Emit an event to the parent.
    */
@@ -118,6 +162,22 @@ export class WorkerRuntime {
         // Respond to ping
         this.send_message('worker.heartbeat', { timestamp: new Date().toISOString() })
         break
+
+      case 'llm.response': {
+        const call_id = payload.call_id as string
+        const pending = this.pending_calls.get(call_id)
+        if (pending) {
+          this.pending_calls.delete(call_id)
+          // Resolve with LLM response
+          const response: LLMResponse = {
+            content: (payload.content as string) || '',
+            usage: payload.usage as LLMResponse['usage'],
+            tool_calls: payload.tool_calls as LLMResponse['tool_calls']
+          }
+          pending.resolve(response as any)
+        }
+        break
+      }
     }
   }