@blockrun/franklin 3.7.9 → 3.7.10

package/dist/agent/llm.d.ts

@@ -27,6 +27,31 @@ export interface LLMClientOptions {
     chain: Chain;
     debug?: boolean;
 }
+/**
+ * Apply Anthropic prompt caching using the `system_and_3` strategy.
+ * Pattern from nousresearch/hermes-agent `agent/prompt_caching.py`.
+ *
+ * Places 4 cache_control breakpoints (Anthropic's max):
+ *   1. System prompt (stable across all turns)
+ *   2-4. Last 3 non-system messages (rolling window)
+ *
+ * Also caches the last tool definition (tools are stable across turns).
+ *
+ * This keeps the cache warm: each new turn extends the cached prefix rather
+ * than invalidating it. Multi-turn conversations see ~75% input token savings
+ * on Anthropic models.
+ */
+/**
+ * True if the given Anthropic model accepts the `thinking: { type: 'enabled' }`
+ * API flag (so-called *extended thinking*). Models using *adaptive thinking*
+ * (Opus 4.7 and later) reject that flag — the behavior is built in and not
+ * opt-in via API. Keeping the allowlist explicit, not derived from a regex,
+ * so a future model that happens to include "opus" in its name doesn't
+ * silently re-enable extended thinking on a model that can't handle it.
+ *
+ * Exported so tests can pin this decision without a live API.
+ */
+export declare function modelHasExtendedThinking(model: string): boolean;
 export declare class ModelClient {
     private apiUrl;
     private chain;

package/dist/agent/llm.js

@@ -20,6 +20,26 @@ import { USER_AGENT } from '../config.js';
  * than invalidating it. Multi-turn conversations see ~75% input token savings
  * on Anthropic models.
  */
+/**
+ * True if the given Anthropic model accepts the `thinking: { type: 'enabled' }`
+ * API flag (so-called *extended thinking*). Models using *adaptive thinking*
+ * (Opus 4.7 and later) reject that flag — the behavior is built in and not
+ * opt-in via API. Keeping the allowlist explicit, not derived from a regex,
+ * so a future model that happens to include "opus" in its name doesn't
+ * silently re-enable extended thinking on a model that can't handle it.
+ *
+ * Exported so tests can pin this decision without a live API.
+ */
+export function modelHasExtendedThinking(model) {
+    const m = model.toLowerCase();
+    if (m.includes('opus-4.7') || m.includes('opus-4-7'))
+        return false;
+    return (m.includes('opus-4.6') || m.includes('opus-4-6') ||
+        m.includes('opus-4.5') || m.includes('opus-4-5') ||
+        m.includes('opus-4.1') || m.includes('opus-4-1') ||
+        m.includes('sonnet-4') ||
+        m.includes('sonnet-3.7'));
+}
 function applyAnthropicPromptCaching(payload, request) {
     const out = { ...payload };
     const cacheMarker = { type: 'ephemeral' };
@@ -140,17 +160,11 @@ export class ModelClient {
         if (isAnthropic) {
             // ─ Anthropic extended thinking ──────────────────────────────────────
             // Enable thinking for Claude models that support it (Opus 4.6, Sonnet 4.6).
-            // This is the single biggest quality lever — Claude with thinking enabled
-            // is dramatically better at complex multi-step tasks, reasoning, and code.
-            //
-            // Uses adaptive thinking: the model decides how much to think per request.
-            // budget_tokens is the MAX it can use (not a minimum), so the model won't
-            // waste tokens on simple tasks. Set to 80% of max_tokens to leave room
-            // for the actual response.
-            const supportsThinking = request.model.includes('opus') ||
-                request.model.includes('sonnet-4') ||
-                request.model.includes('sonnet-3.7');
-            if (supportsThinking) {
+            // Enable the `thinking` API block only for models that accept it.
+            // Claude Opus 4.7 and newer use *adaptive* thinking (built-in, no API
+            // flag); passing the extended-thinking flag to them makes Anthropic
+            // reject the request. See `modelHasExtendedThinking` for the allowlist.
+            if (modelHasExtendedThinking(request.model)) {
                 const maxOut = (request.max_tokens ?? 16_384);
                 requestPayload['thinking'] = {
                     type: 'enabled',

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@blockrun/franklin",
-  "version": "3.7.9",
+  "version": "3.7.10",
   "description": "Franklin — The AI agent with a wallet. Spends USDC autonomously to get real work done. Pay per action, no subscriptions.",
   "type": "module",
   "exports": {