@code-yeongyu/senpi 2026.8.30 → 2026.9.2-3

package/node_modules/@earendil-works/pi-ai/dist/api/anthropic-messages.js

@@ -1,5 +1,6 @@
 import Anthropic from "@anthropic-ai/sdk";
 import { calculateCost } from "../models.js";
+import { registerSessionResourceCleanup } from "../session-resources.js";
 import { isVideoMimeType } from "../types.js";
 import { combineAbortSignals } from "../utils/abort-signals.js";
 import { splitDeferredTools } from "../utils/deferred-tools.js";
@@ -181,6 +182,20 @@ const SERVER_SIDE_FALLBACK_BETA = "server-side-fallback-2026-07-01";
 // A provider can reject its own empty signatures. Learn that capability for one
 // conversation without changing the shared model definition used by other sessions.
 const unsignedThinkingTextReplayFallbacks = new Set();
+registerSessionResourceCleanup((sessionId) => {
+    // One small entry per (session, base URL, model) that ever hit the fallback;
+    // drop the session's entries at teardown so long-lived hosts do not collect
+    // them for every session that ever ran.
+    if (sessionId === undefined) {
+        unsignedThinkingTextReplayFallbacks.clear();
+        return;
+    }
+    const prefix = `${sessionId}\u0000`;
+    for (const key of unsignedThinkingTextReplayFallbacks) {
+        if (key.startsWith(prefix))
+            unsignedThinkingTextReplayFallbacks.delete(key);
+    }
+});
 function unsignedThinkingFallbackKey(model, sessionId) {
     return sessionId === undefined ? undefined : `${sessionId}\u0000${model.baseUrl}\u0000${model.id}`;
 }
@@ -237,12 +252,13 @@ const REPLAYABLE_ANTHROPIC_PROVIDER_NATIVE_TYPES = new Set([
     "text_editor_code_execution_tool_result",
     "tool_search_tool_result",
     "container_upload",
-    // Server-side fallback beta (server-side-fallback-2026-06-01) emits a
-    // `fallback` marker mid-response. The marker itself is replayed as a kept
-    // audit block. Blocks emitted *before* the final marker are the declined
-    // attempt and are pruned in convertMessages (see lastAnthropicFallbackBoundary
-    // and collectDiscardedFallbackToolCallIds); the marker onward replays verbatim.
-    "fallback",
+    // The server-side fallback marker (`fallback`, server-side-fallback-2026-06-01
+    // beta) is deliberately absent: the Messages API rejects it as an *input* tag
+    // ("Input tag 'fallback' found using 'type' does not match any of the expected
+    // tags"), so replaying it 400s every subsequent same-model request. The stored
+    // marker is audit metadata and still drives declined-attempt pruning in
+    // convertMessages (see lastAnthropicFallbackBoundary and
+    // collectDiscardedFallbackToolCallIds); blocks after it replay verbatim.
 ]);
 function isReplayableAnthropicProviderNativeBlock(raw) {
     return isRecord(raw) && typeof raw.type === "string" && REPLAYABLE_ANTHROPIC_PROVIDER_NATIVE_TYPES.has(raw.type);
@@ -1796,7 +1812,8 @@ function convertMessages(transformedMessages, model, isOAuthToken, cacheControl,
             const blocks = [];
             const isSameModel = isSameAnthropicModel(msg, model);
             // Blocks before the final fallback marker are the declined attempt; the
-            // marker onward is the serving model's output and replays verbatim.
+            // blocks after it are the serving model's output and replay verbatim. The
+            // marker itself never replays: the API rejects `fallback` as an input tag.
             const fallbackBoundary = isSameModel ? lastAnthropicFallbackBoundary(msg.content) : -1;
             const preBoundaryPairedServerToolUseIds = fallbackBoundary >= 0
                 ? pairedServerToolUseIdsBeforeBoundary(msg.content, fallbackBoundary)