⚡ Bolt: Add in-memory cache to SearchTool

.jules/bolt.md

@@ -1,3 +1,7 @@
 ## 2026-04-25 - LLM API Connection Pooling
 **Learning:** Making repeated API calls to external LLM providers (e.g. OpenRouter, OpenAI) without HTTP Keep-Alive results in 100-200ms of unnecessary TCP/TLS handshake overhead per request. In an agentic loop, this accumulates significantly.
 **Action:** When initializing HTTP clients (like Axios) for repeated internal/external services, configure them with an `https.Agent` setting `keepAlive: true` to pool connections.
+
+## 2026-05-28 - Agentic Loop Caching Optimization
+**Learning:** In agentic tools that repeatedly query the same context (e.g., `SearchTool.deep`), making redundant operations or API calls causes significant bottlenecks. Caching within the `core.execute()` wrapper ensures standardized formatting, logging, and retry logic are maintained while eliminating the overhead.
+**Action:** Use an in-memory `Map` cache for agentic tools that perform identical context queries during loops, placing cache logic inside the execution wrapper.

src/tools/index.ts

@@ -60,21 +60,32 @@ export class FileTool {
 }
 
 export class SearchTool {
+    // Cache repeated context queries to prevent redundant API calls during agentic loops
+    private cache = new Map<string, any>();
+
     constructor(private core: CoreEngine) {}
 
     async deep(query: string): Promise<ExecutionResult<any>> {
         this.core.log(`Initiating deep search for: ${query}`);
         // This is where real API calls to Google/X/Reddit would go
         return this.core.execute(async () => {
+            if (this.cache.has(query)) {
+                this.core.log(`Cache hit for query: ${query}`);
+                return this.cache.get(query);
+            }
+
             // Simulated results for the SDK base
-            return {
+            const result = {
                 query,
                 timestamp: new Date().toISOString(),
                 findings: [
                     "Best practices for " + query,
                     "Trending solutions in 2026"
                 ]
             };
+
+            this.cache.set(query, result);
+            return result;
         });
     }
 }