@code-yeongyu/senpi-ai 2026.9.8 → 2026.9.9

package/dist/api/anthropic-messages.js

@@ -1618,11 +1618,19 @@ function buildParams(model, context, isOAuthToken, options, unsignedThinkingRepl
             disableThinkingForRequest(params, model, compat);
         }
     }
-    // Anthropic rejects a thinking-enabled request whose final assistant turn
-    // contains tool_use but does not begin with a thinking block. Cross-model
-    // histories lose their signed thinking blocks (demoted to text or dropped),
-    // so degrade thinking for this request instead of failing every turn.
-    if (params.thinking && params.thinking.type !== "disabled" && finalAssistantTurnStartsWithToolUse(params.messages))
+    // A budget-thinking request whose final assistant turn came from another API
+    // and starts with tool_use (its thinking demoted to text or dropped) is the
+    // history shape Anthropic has rejected with "final assistant message must
+    // start with a thinking block", so that request degrades thinking instead of
+    // failing every turn. Nothing else degrades: adaptive families accept a
+    // tool_use-first final turn (verified live 2026-09-09 on claude-opus-5,
+    // claude-opus-4-6 and claude-fable-5), the model itself skips thinking before
+    // trivial tool calls, and every change to `thinking` re-keys the prompt
+    // cache - degrading here re-wrote the whole cached prefix on each tool
+    // continuation (the "cache misses every second prompt" report).
+    if (params.thinking?.type === "enabled" &&
+        finalAssistantTurnIsForeign(context.messages) &&
+        finalAssistantTurnStartsWithToolUse(params.messages))
         disableThinkingForRequest(params, model, compat);
     if (options?.metadata) {
         const userId = options.metadata.user_id;
@@ -1663,6 +1671,16 @@ function applyExtraBodyToAnthropicParams(params, extraBody) {
  * thinking/redacted_thinking block. Anthropic requires thinking-enabled
  * requests to start that turn with a thinking block.
  */
+/** True when the last assistant message was produced through a different wire API. */
+function finalAssistantTurnIsForeign(messages) {
+    for (let index = messages.length - 1; index >= 0; index--) {
+        const message = messages[index];
+        if (message.role !== "assistant")
+            continue;
+        return message.api !== "anthropic-messages";
+    }
+    return false;
+}
 function finalAssistantTurnStartsWithToolUse(messages) {
     for (let index = messages.length - 1; index >= 0; index--) {
         const message = messages[index];