@juspay/neurolink 9.59.1 → 9.59.3

package/dist/core/baseProvider.d.ts

@@ -279,13 +279,20 @@ export declare abstract class BaseProvider implements AIProvider {
      */
     protected validateStreamOptions(options: StreamOptions): void;
     /**
-     * Create text stream transformation - delegated to StreamHandler
+     * Create text stream transformation - delegated to StreamHandler.
+     * Reviewer follow-up: forwards the optional `getUnderlyingError`
+     * callback so providers can capture upstream errors via
+     * `streamText`'s `onError` and have them flow into the
+     * NoOutputGeneratedError sentinel's `providerError` /
+     * `modelResponseRaw`.
      */
     protected createTextStream(result: {
         textStream: AsyncIterable<string>;
-    }): AsyncGenerator<{
+        finishReason?: Promise<unknown> | unknown;
+        totalUsage?: Promise<unknown> | unknown;
+    }, getUnderlyingError?: () => unknown): AsyncGenerator<{
         content: string;
-    }>;
+    } | import("../types/index.js").StreamNoOutputSentinel>;
     /**
      * Create standardized stream result - delegated to StreamHandler
      */

package/dist/core/baseProvider.js

@@ -1019,10 +1019,15 @@ export class BaseProvider {
         this.streamHandler.validateStreamOptions(options);
     }
     /**
-     * Create text stream transformation - delegated to StreamHandler
+     * Create text stream transformation - delegated to StreamHandler.
+     * Reviewer follow-up: forwards the optional `getUnderlyingError`
+     * callback so providers can capture upstream errors via
+     * `streamText`'s `onError` and have them flow into the
+     * NoOutputGeneratedError sentinel's `providerError` /
+     * `modelResponseRaw`.
      */
-    createTextStream(result) {
-        return this.streamHandler.createTextStream(result);
+    createTextStream(result, getUnderlyingError) {
+        return this.streamHandler.createTextStream(result, getUnderlyingError);
     }
     /**
      * Create standardized stream result - delegated to StreamHandler

package/dist/core/modules/StreamHandler.d.ts

@@ -12,7 +12,7 @@
  *
  * @module core/modules/StreamHandler
  */
-import type { StreamOptions, StreamResult, UnknownRecord, AIProviderName } from "../../types/index.js";
+import type { StreamOptions, StreamResult, UnknownRecord, AIProviderName, StreamNoOutputSentinel } from "../../types/index.js";
 /**
  * StreamHandler class - Handles streaming operations for AI providers
  */
@@ -30,9 +30,28 @@ export declare class StreamHandler {
      */
     createTextStream(result: {
         textStream: AsyncIterable<string>;
-    }): AsyncGenerator<{
+        /**
+         * Optional metadata getters from the AI SDK's StreamTextResult. These
+         * reject with NoOutputGeneratedError when no output is produced, which
+         * is exactly the path Curator's P3-6 fix needs to enrich. We attempt
+         * to await them in the catch block; whichever resolve get included in
+         * the sentinel chunk metadata.
+         */
+        finishReason?: Promise<unknown> | unknown;
+        totalUsage?: Promise<unknown> | unknown;
+    }, 
+    /**
+     * Reviewer follow-up: optional getter for the provider's captured
+     * upstream error (typically wired from `streamText`'s `onError`
+     * callback). When set, the sentinel's `providerError` /
+     * `modelResponseRaw` reflect the real upstream cause instead of the
+     * AI SDK's generic "No output generated" message. Callers that don't
+     * capture upstream errors can omit this — the sentinel still
+     * populates with the AI SDK error.
+     */
+    getUnderlyingError?: () => unknown): AsyncGenerator<{
         content: string;
-    }>;
+    } | StreamNoOutputSentinel>;
     /**
      * Create standardized stream result - consolidates result structure
      */

package/dist/core/modules/StreamHandler.js

@@ -16,6 +16,7 @@ import { trace, context as otelContext, SpanStatusCode, } from "@opentelemetry/a
 import { tracers, ATTR, withSpan } from "../../telemetry/index.js";
 import { logger } from "../../utils/logger.js";
 import { validateStreamOptions as validateStreamOpts, ValidationError, createValidationSummary, } from "../../utils/parameterValidation.js";
+import { buildNoOutputSentinel, detectPostStreamNoOutput, stampNoOutputSpan, } from "../../utils/noOutputSentinel.js";
 import { STEP_LIMITS } from "../constants.js";
 import { createAnalytics } from "../analytics.js";
 import { nanoid } from "nanoid";
@@ -83,7 +84,17 @@ export class StreamHandler {
      * Create text stream transformation - consolidates identical logic from 7/10 providers
      * Tracks TTFC (Time To First Chunk), chunk count, and total bytes streamed.
      */
-    createTextStream(result) {
+    createTextStream(result, 
+    /**
+     * Reviewer follow-up: optional getter for the provider's captured
+     * upstream error (typically wired from `streamText`'s `onError`
+     * callback). When set, the sentinel's `providerError` /
+     * `modelResponseRaw` reflect the real upstream cause instead of the
+     * AI SDK's generic "No output generated" message. Callers that don't
+     * capture upstream errors can omit this — the sentinel still
+     * populates with the AI SDK error.
+     */
+    getUnderlyingError) {
         const providerName = this.providerName;
         return (async function* () {
             let chunkCount = 0;
@@ -113,32 +124,43 @@ export class StreamHandler {
                 // rather than crashing the process with an unhandled rejection.
                 if (NoOutputGeneratedError.isInstance(error)) {
                     logger.warn(`${providerName}: Stream produced no output (NoOutputGeneratedError), returning empty stream`);
-                    // Curator P2-5: stamp the active OTel span so ContextEnricher.onEnd()
-                    // surfaces a WARNING-level Langfuse observation instead of defaulting
-                    // to DEFAULT with no status message.
-                    try {
-                        const activeSpan = trace.getSpan(otelContext.active());
-                        if (activeSpan) {
-                            activeSpan.setAttribute("neurolink.no_output", true);
-                        }
-                    }
-                    catch {
-                        // Tracing not initialized — ignore.
-                    }
+                    // Curator P3-6: build the enriched sentinel using the shared
+                    // helper so every provider yields the same shape. Pass the
+                    // captured upstream error (if any) so providerError /
+                    // modelResponseRaw carry the real cause.
+                    const sentinel = await buildNoOutputSentinel(error, result, getUnderlyingError?.());
+                    // Curator P2-5 + P3-6: stamp the active OTel span so
+                    // ContextEnricher.onEnd() surfaces a WARNING-level Langfuse
+                    // observation with finishReason + token usage. Centralized in
+                    // stampNoOutputSpan so every wired site stamps consistently.
+                    stampNoOutputSpan(sentinel);
                     // S4 fix: yield a sentinel chunk so Pipeline B can detect the empty stream
                     // and set the span to WARNING status instead of OK
-                    yield {
-                        content: "",
-                        metadata: {
-                            noOutput: true,
-                            errorType: "NoOutputGeneratedError",
-                        },
-                    };
+                    yield sentinel;
+                    // Reviewer follow-up: must return here. Falling through to the
+                    // post-stream detection block below would yield a SECOND sentinel
+                    // chunk (verified with synthetic NoOutputGeneratedError stream:
+                    // count=2 sentinels). The catch block's yield is sufficient.
+                    return;
                 }
                 else {
                     throw error;
                 }
             }
+            // Curator P3-6 (round-2 fix): the production trigger sets
+            // NoOutputGeneratedError on `result.finishReason` rejection — NOT
+            // thrown from textStream iteration. Surface that path here so the
+            // sentinel actually fires for real-world no-output streams. The
+            // catch above remains as a defensive path for failure modes that
+            // do throw from textStream.
+            if (chunkCount === 0) {
+                const detected = await detectPostStreamNoOutput(result, getUnderlyingError?.());
+                if (detected) {
+                    logger.warn(`${providerName}: Stream produced no output (NoOutputGeneratedError) — caught from finishReason rejection`);
+                    stampNoOutputSpan(detected.sentinel);
+                    yield detected.sentinel;
+                }
+            }
             // Record completion metrics on the active span
             const activeSpan = trace.getSpan(otelContext.active());
             if (activeSpan) {

package/dist/lib/core/baseProvider.d.ts

@@ -279,13 +279,20 @@ export declare abstract class BaseProvider implements AIProvider {
      */
     protected validateStreamOptions(options: StreamOptions): void;
     /**
-     * Create text stream transformation - delegated to StreamHandler
+     * Create text stream transformation - delegated to StreamHandler.
+     * Reviewer follow-up: forwards the optional `getUnderlyingError`
+     * callback so providers can capture upstream errors via
+     * `streamText`'s `onError` and have them flow into the
+     * NoOutputGeneratedError sentinel's `providerError` /
+     * `modelResponseRaw`.
      */
     protected createTextStream(result: {
         textStream: AsyncIterable<string>;
-    }): AsyncGenerator<{
+        finishReason?: Promise<unknown> | unknown;
+        totalUsage?: Promise<unknown> | unknown;
+    }, getUnderlyingError?: () => unknown): AsyncGenerator<{
         content: string;
-    }>;
+    } | import("../types/index.js").StreamNoOutputSentinel>;
     /**
      * Create standardized stream result - delegated to StreamHandler
      */

package/dist/lib/core/baseProvider.js

@@ -1019,10 +1019,15 @@ export class BaseProvider {
         this.streamHandler.validateStreamOptions(options);
     }
     /**
-     * Create text stream transformation - delegated to StreamHandler
+     * Create text stream transformation - delegated to StreamHandler.
+     * Reviewer follow-up: forwards the optional `getUnderlyingError`
+     * callback so providers can capture upstream errors via
+     * `streamText`'s `onError` and have them flow into the
+     * NoOutputGeneratedError sentinel's `providerError` /
+     * `modelResponseRaw`.
      */
-    createTextStream(result) {
-        return this.streamHandler.createTextStream(result);
+    createTextStream(result, getUnderlyingError) {
+        return this.streamHandler.createTextStream(result, getUnderlyingError);
     }
     /**
      * Create standardized stream result - delegated to StreamHandler

package/dist/lib/core/modules/StreamHandler.d.ts

@@ -12,7 +12,7 @@
  *
  * @module core/modules/StreamHandler
  */
-import type { StreamOptions, StreamResult, UnknownRecord, AIProviderName } from "../../types/index.js";
+import type { StreamOptions, StreamResult, UnknownRecord, AIProviderName, StreamNoOutputSentinel } from "../../types/index.js";
 /**
  * StreamHandler class - Handles streaming operations for AI providers
  */
@@ -30,9 +30,28 @@ export declare class StreamHandler {
      */
     createTextStream(result: {
         textStream: AsyncIterable<string>;
-    }): AsyncGenerator<{
+        /**
+         * Optional metadata getters from the AI SDK's StreamTextResult. These
+         * reject with NoOutputGeneratedError when no output is produced, which
+         * is exactly the path Curator's P3-6 fix needs to enrich. We attempt
+         * to await them in the catch block; whichever resolve get included in
+         * the sentinel chunk metadata.
+         */
+        finishReason?: Promise<unknown> | unknown;
+        totalUsage?: Promise<unknown> | unknown;
+    }, 
+    /**
+     * Reviewer follow-up: optional getter for the provider's captured
+     * upstream error (typically wired from `streamText`'s `onError`
+     * callback). When set, the sentinel's `providerError` /
+     * `modelResponseRaw` reflect the real upstream cause instead of the
+     * AI SDK's generic "No output generated" message. Callers that don't
+     * capture upstream errors can omit this — the sentinel still
+     * populates with the AI SDK error.
+     */
+    getUnderlyingError?: () => unknown): AsyncGenerator<{
         content: string;
-    }>;
+    } | StreamNoOutputSentinel>;
     /**
      * Create standardized stream result - consolidates result structure
      */

package/dist/lib/core/modules/StreamHandler.js

@@ -16,6 +16,7 @@ import { trace, context as otelContext, SpanStatusCode, } from "@opentelemetry/a
 import { tracers, ATTR, withSpan } from "../../telemetry/index.js";
 import { logger } from "../../utils/logger.js";
 import { validateStreamOptions as validateStreamOpts, ValidationError, createValidationSummary, } from "../../utils/parameterValidation.js";
+import { buildNoOutputSentinel, detectPostStreamNoOutput, stampNoOutputSpan, } from "../../utils/noOutputSentinel.js";
 import { STEP_LIMITS } from "../constants.js";
 import { createAnalytics } from "../analytics.js";
 import { nanoid } from "nanoid";
@@ -83,7 +84,17 @@ export class StreamHandler {
      * Create text stream transformation - consolidates identical logic from 7/10 providers
      * Tracks TTFC (Time To First Chunk), chunk count, and total bytes streamed.
      */
-    createTextStream(result) {
+    createTextStream(result, 
+    /**
+     * Reviewer follow-up: optional getter for the provider's captured
+     * upstream error (typically wired from `streamText`'s `onError`
+     * callback). When set, the sentinel's `providerError` /
+     * `modelResponseRaw` reflect the real upstream cause instead of the
+     * AI SDK's generic "No output generated" message. Callers that don't
+     * capture upstream errors can omit this — the sentinel still
+     * populates with the AI SDK error.
+     */
+    getUnderlyingError) {
         const providerName = this.providerName;
         return (async function* () {
             let chunkCount = 0;
@@ -113,32 +124,43 @@ export class StreamHandler {
                 // rather than crashing the process with an unhandled rejection.
                 if (NoOutputGeneratedError.isInstance(error)) {
                     logger.warn(`${providerName}: Stream produced no output (NoOutputGeneratedError), returning empty stream`);
-                    // Curator P2-5: stamp the active OTel span so ContextEnricher.onEnd()
-                    // surfaces a WARNING-level Langfuse observation instead of defaulting
-                    // to DEFAULT with no status message.
-                    try {
-                        const activeSpan = trace.getSpan(otelContext.active());
-                        if (activeSpan) {
-                            activeSpan.setAttribute("neurolink.no_output", true);
-                        }
-                    }
-                    catch {
-                        // Tracing not initialized — ignore.
-                    }
+                    // Curator P3-6: build the enriched sentinel using the shared
+                    // helper so every provider yields the same shape. Pass the
+                    // captured upstream error (if any) so providerError /
+                    // modelResponseRaw carry the real cause.
+                    const sentinel = await buildNoOutputSentinel(error, result, getUnderlyingError?.());
+                    // Curator P2-5 + P3-6: stamp the active OTel span so
+                    // ContextEnricher.onEnd() surfaces a WARNING-level Langfuse
+                    // observation with finishReason + token usage. Centralized in
+                    // stampNoOutputSpan so every wired site stamps consistently.
+                    stampNoOutputSpan(sentinel);
                     // S4 fix: yield a sentinel chunk so Pipeline B can detect the empty stream
                     // and set the span to WARNING status instead of OK
-                    yield {
-                        content: "",
-                        metadata: {
-                            noOutput: true,
-                            errorType: "NoOutputGeneratedError",
-                        },
-                    };
+                    yield sentinel;
+                    // Reviewer follow-up: must return here. Falling through to the
+                    // post-stream detection block below would yield a SECOND sentinel
+                    // chunk (verified with synthetic NoOutputGeneratedError stream:
+                    // count=2 sentinels). The catch block's yield is sufficient.
+                    return;
                 }
                 else {
                     throw error;
                 }
             }
+            // Curator P3-6 (round-2 fix): the production trigger sets
+            // NoOutputGeneratedError on `result.finishReason` rejection — NOT
+            // thrown from textStream iteration. Surface that path here so the
+            // sentinel actually fires for real-world no-output streams. The
+            // catch above remains as a defensive path for failure modes that
+            // do throw from textStream.
+            if (chunkCount === 0) {
+                const detected = await detectPostStreamNoOutput(result, getUnderlyingError?.());
+                if (detected) {
+                    logger.warn(`${providerName}: Stream produced no output (NoOutputGeneratedError) — caught from finishReason rejection`);
+                    stampNoOutputSpan(detected.sentinel);
+                    yield detected.sentinel;
+                }
+            }
             // Record completion metrics on the active span
             const activeSpan = trace.getSpan(otelContext.active());
             if (activeSpan) {