@code-yeongyu/senpi-ai 2026.10.5 → 2026.10.6

package/dist/utils/overflow.d.ts

@@ -67,6 +67,7 @@ export declare function isRecoverableLength(message: AssistantMessage, desiredMa
  */
 export declare function getOverflowPatterns(): RegExp[];
 export declare function isCursorPayloadResourceExhausted(message: {
+    provider?: string;
     stopReason?: string;
     errorMessage?: string;
     usage?: {
@@ -83,6 +84,7 @@ export declare function isCursorPayloadResourceExhausted(message: {
  * context window.
  */
 export declare function isCursorQuotaResourceExhausted(message: {
+    provider?: string;
     stopReason?: string;
     errorMessage?: string;
     usage?: {
@@ -94,6 +96,7 @@ export declare function isCursorQuotaResourceExhausted(message: {
     };
 }, contextWindow: number): boolean;
 export declare function isCursorZeroTokenResourceExhausted(message: {
+    provider?: string;
     stopReason?: string;
     errorMessage?: string;
     usage?: {

package/dist/utils/overflow.js

@@ -215,6 +215,15 @@ export function isRecoverableLength(message, desiredMaxOutput) {
 export function getOverflowPatterns() {
     return [...OVERFLOW_PATTERNS];
 }
+const CURSOR_PROVIDERS = new Set(["cursor", "cursor-cli-oauth"]);
+/**
+ * The zero-token / quota `resource_exhausted` signatures below describe Cursor's backend only.
+ * Other providers send `resource_exhausted` for their own rate and usage limits (senpi#2660), and
+ * reading those as a Cursor payload overflow or re-mint skips the fallback chain.
+ */
+function isCursorProviderMessage(message) {
+    return message.provider === undefined || CURSOR_PROVIDERS.has(message.provider);
+}
 function cursorZeroTokenCount(message) {
     const usage = message.usage;
     if (!usage)
@@ -222,6 +231,8 @@ function cursorZeroTokenCount(message) {
     return (usage.totalTokens || (usage.input ?? 0) + (usage.output ?? 0) + (usage.cacheRead ?? 0) + (usage.cacheWrite ?? 0));
 }
 export function isCursorPayloadResourceExhausted(message, _estimateTokens) {
+    if (!isCursorProviderMessage(message))
+        return false;
     if (message.stopReason !== "error" || !/resource.?exhausted/i.test(message.errorMessage || "")) {
         return false;
     }
@@ -234,13 +245,16 @@ export function isCursorPayloadResourceExhausted(message, _estimateTokens) {
  */
 export function isCursorQuotaResourceExhausted(message, contextWindow) {
     const tokens = cursorZeroTokenCount(message);
-    return (message.stopReason === "error" &&
+    return (isCursorProviderMessage(message) &&
+        message.stopReason === "error" &&
         RESOURCE_EXHAUSTED_PATTERN.test(message.errorMessage || "") &&
         contextWindow > 0 &&
         tokens > 0 &&
         tokens < contextWindow * 0.5);
 }
 export function isCursorZeroTokenResourceExhausted(message) {
+    if (!isCursorProviderMessage(message))
+        return false;
     if (message.stopReason !== "error" || !/resource.?exhausted/i.test(message.errorMessage || "")) {
         return false;
     }

package/dist/utils/retry-hint.d.ts

@@ -6,7 +6,7 @@
  *   2. retry-after header      (strict integer delta-seconds `^\d+$`, else HTTP-date)
  *   3. x-ratelimit-reset{,-requests,-tokens}  (epoch seconds, max-wins)
  *   4. JSON retryDelay strings (ms/s/m units, recursive, max-wins; numeric rejected)
- *   5. body prose              (ms-marker, seconds-marker, relative, resets-at ISO8601)
+ *   5. body prose              (ms-marker, seconds-marker, relative, resets-in, resets-at ISO8601)
  *
  * Eligibility gate: status 429 OR body marks rate-limit error.
  * Explicit zero = 0 (beats lower-precedence positives).

package/dist/utils/retry-hint.js

@@ -6,7 +6,7 @@
  *   2. retry-after header      (strict integer delta-seconds `^\d+$`, else HTTP-date)
  *   3. x-ratelimit-reset{,-requests,-tokens}  (epoch seconds, max-wins)
  *   4. JSON retryDelay strings (ms/s/m units, recursive, max-wins; numeric rejected)
- *   5. body prose              (ms-marker, seconds-marker, relative, resets-at ISO8601)
+ *   5. body prose              (ms-marker, seconds-marker, relative, resets-in, resets-at ISO8601)
  *
  * Eligibility gate: status 429 OR body marks rate-limit error.
  * Explicit zero = 0 (beats lower-precedence positives).
@@ -132,6 +132,10 @@ function fromBodyProse(bodyText, nowMs) {
         return Number.parseInt(m[1], 10) * 1000;
     // relative prose: try again | retry | wait after ... [in] N <unit>
     m = bodyText.match(/(?:try again|retry|wait after).*?(?:in\s+)?(\d+(?:\.\d+)?)\s*(ms|s|sec|seconds?|m|min|minutes?)\b/i);
+    if (m)
+        return convertUnit(Number.parseFloat(m[1]), m[2]);
+    // limit prose: [limit will] reset(s) in N <unit>
+    m = bodyText.match(/\bresets?\s+in\s+(\d+(?:\.\d+)?)\s*(ms|s|sec|seconds?|m|min|minutes?)\b/i);
     if (m)
         return convertUnit(Number.parseFloat(m[1]), m[2]);
     // resets at <ISO8601-with-tz>

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@code-yeongyu/senpi-ai",
-	"version": "2026.10.5",
+	"version": "2026.10.6",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"type": "module",
 	"main": "./dist/index.js",
@@ -88,7 +88,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "0.127.0",
 		"@aws-sdk/client-bedrock-runtime": "3.1136.0",
-		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.10.5",
+		"@earendil-works/pi-telemetry": "npm:@code-yeongyu/senpi-telemetry@2026.10.6",
 		"@google/genai": "2.23.0",
 		"@smithy/node-http-handler": "4.12.1",
 		"http-proxy-agent": "9.1.0",