@juspay/neurolink 9.10.0 → 9.11.0

package/CHANGELOG.md

@@ -1,3 +1,15 @@
+## [9.11.0](https://github.com/juspay/neurolink/compare/v9.10.1...v9.11.0) (2026-02-22)
+
+### Features
+
+- **(core):** provider error standardization, abort hardening, session budget, and continuous test suites ([0ceb590](https://github.com/juspay/neurolink/commit/0ceb590dcad315d392f3e9563d0c5dc0c83cda00))
+
+## [9.10.1](https://github.com/juspay/neurolink/compare/v9.10.0...v9.10.1) (2026-02-21)
+
+### Bug Fixes
+
+- **(video-analysis):** add stream support for video analysis ([938aeef](https://github.com/juspay/neurolink/commit/938aeef876277360700d2a7192155af1f1316f28))
+
 ## [9.10.0](https://github.com/juspay/neurolink/compare/v9.9.0...v9.10.0) (2026-02-20)
 
 ### Features

package/dist/adapters/video/videoAnalyzer.d.ts

@@ -8,16 +8,16 @@
  */
 import { AIProviderName } from "../../constants/enums.js";
 import type { CoreMessage } from "ai";
-export declare function analyzeVideoWithVertexAI(frames: CoreMessage, options?: {
+export declare function analyzeVideoWithVertexAI(messages: CoreMessage[], options?: {
     project?: string;
     location?: string;
     model?: string;
 }): Promise<string>;
-export declare function analyzeVideoWithGeminiAPI(frames: CoreMessage, options?: {
+export declare function analyzeVideoWithGeminiAPI(messages: CoreMessage[], options?: {
     apiKey?: string;
     model?: string;
 }): Promise<string>;
-export declare function analyzeVideo(frames: CoreMessage, options?: {
+export declare function analyzeVideo(messages: CoreMessage[], options?: {
     provider?: AIProviderName;
     project?: string;
     location?: string;

package/dist/adapters/video/videoAnalyzer.js

@@ -9,23 +9,35 @@
 import { AIProviderName, ErrorSeverity, ErrorCategory, } from "../../constants/enums.js";
 import { logger } from "../../utils/logger.js";
 import { readFile } from "node:fs/promises";
-import { NeuroLinkError } from "../../utils/errorHandling.js";
+import { NeuroLinkError, ErrorFactory } from "../../utils/errorHandling.js";
 // ---------------------------------------------------------------------------
 // Shared config
 // ---------------------------------------------------------------------------
 const DEFAULT_MODEL = "gemini-2.0-flash";
 const DEFAULT_LOCATION = "us-central1";
+/**
+ * Extract content items from user messages
+ *
+ * @param messages - Array of CoreMessage objects
+ * @returns Flattened array of content items from user messages
+ */
+function extractUserContent(messages) {
+    const userMessages = messages.filter((msg) => msg.role === "user");
+    return userMessages.flatMap((msg) => Array.isArray(msg.content) ? msg.content : []);
+}
 /**
  * Convert CoreMessage content array to Gemini parts format
  *
- * @param contentArray - Array of content items from CoreMessage
+ * @param messages - Array of CoreMessage objects
  * @returns Array of parts in Gemini API format
  */
-function buildContentParts(frames) {
-    const contentArray = Array.isArray(frames.content) ? frames.content : [];
-    return contentArray.map((item) => {
-        if (item.type === "text" && item.text) {
-            return { text: item.text };
+function buildContentParts(messages) {
+    const allContent = extractUserContent(messages);
+    return allContent
+        .map((item) => {
+        if (item.type === "text") {
+            // Accept text parts regardless of whether text is empty
+            return { text: item.text || "" };
         }
         else if (item.type === "image" && item.image) {
             let base64Data;
@@ -38,7 +50,7 @@ function buildContentParts(frames) {
                 base64Data = item.image.replace(/^data:image\/[a-z]+;base64,/, "");
             }
             else {
-                throw new Error(`Invalid image data type: expected string, Buffer, or Uint8Array, got ${typeof item.image}`);
+                throw ErrorFactory.invalidConfiguration("image data type", `expected string, Buffer, or Uint8Array, got ${typeof item.image}`, { itemType: item.type, dataType: typeof item.image });
             }
             return {
                 inlineData: {
@@ -47,8 +59,14 @@ function buildContentParts(frames) {
                 },
             };
         }
-        throw new Error(`Invalid content type: ${item.type}`);
-    });
+        else if (item.type === "file") {
+            // Skip file parts - not supported in Gemini parts format
+            return null;
+        }
+        // Return null for unsupported types
+        return null;
+    })
+        .filter((part) => part !== null);
 }
 /**
  * Configuration for video frame analysis.
@@ -88,7 +106,7 @@ Ensure the final response is fully self-sufficient and does not reference extern
 // ---------------------------------------------------------------------------
 // Vertex AI
 // ---------------------------------------------------------------------------
-export async function analyzeVideoWithVertexAI(frames, options = {}) {
+export async function analyzeVideoWithVertexAI(messages, options = {}) {
     const startTime = Date.now();
     const { GoogleGenAI } = await import("@google/genai");
     // Get default config and merge with provided options
@@ -96,9 +114,9 @@ export async function analyzeVideoWithVertexAI(frames, options = {}) {
     const project = options.project ?? config.project;
     const location = options.location ?? config.location;
     const model = options.model || DEFAULT_MODEL;
-    // Extract content array from CoreMessage
-    const contentArray = Array.isArray(frames.content) ? frames.content : [];
-    const frameCount = contentArray.filter((item) => item.type === "image").length;
+    // Convert frames content to parts array for Gemini
+    const parts = buildContentParts(messages);
+    const frameCount = parts.filter((part) => "inlineData" in part && part.inlineData).length;
     logger.debug("[GeminiVideoAnalyzer] Analyzing video with Vertex AI", {
         project,
         location,
@@ -106,8 +124,6 @@ export async function analyzeVideoWithVertexAI(frames, options = {}) {
         frameCount,
     });
     const ai = new GoogleGenAI({ vertexai: true, project, location });
-    // Convert frames content to parts array for Gemini
-    const parts = buildContentParts(frames);
     const response = await ai.models.generateContent({
         model,
         config: buildConfig(),
@@ -129,7 +145,7 @@ export async function analyzeVideoWithVertexAI(frames, options = {}) {
 // ---------------------------------------------------------------------------
 // Gemini API (Google AI)
 // ---------------------------------------------------------------------------
-export async function analyzeVideoWithGeminiAPI(frames, options = {}) {
+export async function analyzeVideoWithGeminiAPI(messages, options = {}) {
     const startTime = Date.now();
     const { GoogleGenAI } = await import("@google/genai");
     const apiKey = options.apiKey || process.env.GOOGLE_AI_API_KEY;
@@ -137,16 +153,14 @@ export async function analyzeVideoWithGeminiAPI(frames, options = {}) {
     if (!apiKey) {
         throw new Error("GOOGLE_AI_API_KEY environment variable is required for Gemini API video analysis");
     }
-    // Extract content array from CoreMessage
-    const contentArray = Array.isArray(frames.content) ? frames.content : [];
-    const frameCount = contentArray.filter((item) => item.type === "image").length;
+    // Convert frames content to parts array for Gemini
+    const parts = buildContentParts(messages);
+    const frameCount = parts.filter((part) => "inlineData" in part && part.inlineData).length;
     logger.debug("[GeminiVideoAnalyzer] Analyzing video with Gemini API", {
         model,
         frameCount,
     });
     const ai = new GoogleGenAI({ apiKey });
-    // Convert frames content to parts array for Gemini
-    const parts = buildContentParts(frames);
     logger.debug("[GeminiVideoAnalyzer] Generating analysis with frames");
     const response = await ai.models.generateContent({
         model,
@@ -207,15 +221,15 @@ async function getVertexConfig() {
     }
     return { project, location };
 }
-export async function analyzeVideo(frames, options = {}) {
+export async function analyzeVideo(messages, options = {}) {
     const provider = options.provider || AIProviderName.AUTO;
     // Vertex — only when GOOGLE_VERTEX_PROJECT is explicitly set
     if (provider === AIProviderName.VERTEX || provider === AIProviderName.AUTO) {
-        return analyzeVideoWithVertexAI(frames, options);
+        return analyzeVideoWithVertexAI(messages, options);
     }
     // Gemini API — when GOOGLE_AI_API_KEY is set
     if (provider === AIProviderName.GOOGLE_AI && process.env.GOOGLE_AI_API_KEY) {
-        return analyzeVideoWithGeminiAPI(frames, options);
+        return analyzeVideoWithGeminiAPI(messages, options);
     }
     throw new Error("No valid provider configuration found. " +
         "Set GOOGLE_VERTEX_PROJECT for Vertex AI or GOOGLE_AI_API_KEY for Gemini API.");

package/dist/agent/directTools.d.ts

@@ -219,12 +219,12 @@ export declare const directAgentTools: {
         content: z.ZodString;
         mode: z.ZodDefault<z.ZodEnum<["create", "overwrite", "append"]>>;
     }, "strip", z.ZodTypeAny, {
-        content: string;
         path: string;
+        content: string;
         mode: "append" | "create" | "overwrite";
     }, {
-        content: string;
         path: string;
+        content: string;
         mode?: "append" | "create" | "overwrite" | undefined;
     }>, {
         success: boolean;
@@ -249,8 +249,8 @@ export declare const directAgentTools: {
         written?: undefined;
     }> & {
         execute: (args: {
-            content: string;
             path: string;
+            content: string;
             mode: "append" | "create" | "overwrite";
         }, options: import("ai").ToolExecutionOptions) => PromiseLike<{
             success: boolean;

package/dist/cli/commands/config.d.ts

@@ -60,8 +60,8 @@ declare const ConfigSchema: z.ZodObject<{
         }, {
             model?: string | undefined;
             projectId?: string | undefined;
-            credentials?: string | undefined;
             location?: string | undefined;
+            credentials?: string | undefined;
             privateKey?: string | undefined;
             serviceAccountKey?: string | undefined;
             clientEmail?: string | undefined;
@@ -83,13 +83,13 @@ declare const ConfigSchema: z.ZodObject<{
             model: z.ZodDefault<z.ZodString>;
         }, "strip", z.ZodTypeAny, {
             model: string;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         }, {
             model?: string | undefined;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         }>>;
         "google-ai": z.ZodOptional<z.ZodObject<{
@@ -163,8 +163,8 @@ declare const ConfigSchema: z.ZodObject<{
         } | undefined;
         azure?: {
             model: string;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         } | undefined;
         "google-ai"?: {
@@ -200,8 +200,8 @@ declare const ConfigSchema: z.ZodObject<{
         vertex?: {
             model?: string | undefined;
             projectId?: string | undefined;
-            credentials?: string | undefined;
             location?: string | undefined;
+            credentials?: string | undefined;
             privateKey?: string | undefined;
             serviceAccountKey?: string | undefined;
             clientEmail?: string | undefined;
@@ -212,8 +212,8 @@ declare const ConfigSchema: z.ZodObject<{
         } | undefined;
         azure?: {
             model?: string | undefined;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         } | undefined;
         "google-ai"?: {
@@ -483,8 +483,8 @@ declare const ConfigSchema: z.ZodObject<{
         } | undefined;
         azure?: {
             model: string;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         } | undefined;
         "google-ai"?: {
@@ -569,8 +569,8 @@ declare const ConfigSchema: z.ZodObject<{
         vertex?: {
             model?: string | undefined;
             projectId?: string | undefined;
-            credentials?: string | undefined;
             location?: string | undefined;
+            credentials?: string | undefined;
             privateKey?: string | undefined;
             serviceAccountKey?: string | undefined;
             clientEmail?: string | undefined;
@@ -581,8 +581,8 @@ declare const ConfigSchema: z.ZodObject<{
         } | undefined;
         azure?: {
             model?: string | undefined;
-            endpoint?: string | undefined;
             apiKey?: string | undefined;
+            endpoint?: string | undefined;
             deploymentId?: string | undefined;
         } | undefined;
         "google-ai"?: {

package/dist/cli/loop/optionsSchema.d.ts

@@ -5,4 +5,4 @@ import type { TextGenerationOptions } from "../../lib/types/generateTypes.js";
  * This object provides metadata for validation and help text in the CLI loop.
  * It is derived from the main TextGenerationOptions interface to ensure consistency.
  */
-export declare const textGenerationOptionsSchema: Record<keyof Omit<TextGenerationOptions, "prompt" | "input" | "schema" | "tools" | "context" | "conversationHistory" | "conversationMessages" | "conversationMemoryConfig" | "originalPrompt" | "middleware" | "expectedOutcome" | "evaluationCriteria" | "region" | "csvOptions" | "tts" | "thinkingConfig" | "fileRegistry" | "abortSignal" | "toolFilter" | "excludeTools" | "toolChoice" | "prepareStep">, OptionSchema>;
+export declare const textGenerationOptionsSchema: Record<keyof Omit<TextGenerationOptions, "prompt" | "input" | "schema" | "tools" | "context" | "conversationHistory" | "conversationMessages" | "conversationMemoryConfig" | "originalPrompt" | "middleware" | "expectedOutcome" | "evaluationCriteria" | "region" | "csvOptions" | "tts" | "thinkingConfig" | "requestId" | "fileRegistry" | "abortSignal" | "toolFilter" | "excludeTools" | "toolChoice" | "prepareStep">, OptionSchema>;

package/dist/constants/contextWindows.d.ts

@@ -25,9 +25,12 @@ export declare const MODEL_CONTEXT_WINDOWS: Record<string, Record<string, number
  * Resolve context window size for a provider/model combination.
  *
  * Priority:
- *  1. Exact model match under provider
- *  2. Provider's _default
- *  3. Global DEFAULT_CONTEXT_WINDOW
+ *  0. Dynamic model registry (DynamicModelProvider) — resolves cross-provider
+ *     models (e.g. Claude on Vertex) that the static table cannot handle
+ *  1. Exact model match under provider in static registry
+ *  2. Prefix match under provider in static registry
+ *  3. Provider's _default in static registry
+ *  4. Global DEFAULT_CONTEXT_WINDOW
  */
 export declare function getContextWindowSize(provider: string, model?: string): number;
 /**

package/dist/constants/contextWindows.js

@@ -10,6 +10,8 @@
  * - Google: https://ai.google.dev/gemini-api/docs/models
  * - Others: Provider documentation as of Feb 2026
  */
+import { DynamicModelProvider } from "../core/dynamicModels.js";
+import { logger } from "../utils/logger.js";
 /** Default context window when provider/model is unknown */
 export const DEFAULT_CONTEXT_WINDOW = 128_000;
 /** Maximum output reserve when maxTokens not specified */
@@ -67,6 +69,10 @@ export const MODEL_CONTEXT_WINDOWS = {
         "gemini-2.0-flash": 1_048_576,
         "gemini-1.5-pro": 2_097_152,
         "gemini-1.5-flash": 1_048_576,
+        "claude-sonnet-4-5": 200_000,
+        "claude-sonnet-4-20250514": 200_000,
+        "claude-opus-4": 200_000,
+        "claude-opus-4-20250514": 200_000,
     },
     bedrock: {
         _default: 200_000,
@@ -109,11 +115,32 @@ export const MODEL_CONTEXT_WINDOWS = {
  * Resolve context window size for a provider/model combination.
  *
  * Priority:
- *  1. Exact model match under provider
- *  2. Provider's _default
- *  3. Global DEFAULT_CONTEXT_WINDOW
+ *  0. Dynamic model registry (DynamicModelProvider) — resolves cross-provider
+ *     models (e.g. Claude on Vertex) that the static table cannot handle
+ *  1. Exact model match under provider in static registry
+ *  2. Prefix match under provider in static registry
+ *  3. Provider's _default in static registry
+ *  4. Global DEFAULT_CONTEXT_WINDOW
  */
 export function getContextWindowSize(provider, model) {
+    // Step 0: Check dynamic model registry first.
+    // This resolves cases where the runtime provider differs from the model's
+    // origin (e.g. Claude running via Vertex would hit Vertex's Gemini default
+    // in the static table). The dynamic registry knows the actual model metadata.
+    if (model) {
+        try {
+            const dynamicProvider = DynamicModelProvider.getInstance();
+            const modelConfig = dynamicProvider.resolveModel(provider, model);
+            if (modelConfig?.contextWindow) {
+                logger.debug(`[ContextWindow] Resolved via dynamic registry: provider=${provider}, model=${model}, contextWindow=${modelConfig.contextWindow}`);
+                return modelConfig.contextWindow;
+            }
+        }
+        catch {
+            // Dynamic registry not initialized yet — fall through to static lookup
+        }
+    }
+    // Static fallback chain
     const providerWindows = MODEL_CONTEXT_WINDOWS[provider];
     if (!providerWindows) {
         return DEFAULT_CONTEXT_WINDOW;

package/dist/constants/index.d.ts

@@ -58,7 +58,7 @@ export declare const PROVIDER_OPERATION_CONFIGS: {
     };
     readonly OLLAMA: {
         readonly timeout: 10000;
-        readonly maxRetries: 2;
+        readonly maxRetries: 1;
         readonly retryDelay: 200;
     };
 };
@@ -83,7 +83,7 @@ export declare const MCP_OPERATION_CONFIGS: {
     };
     readonly HEALTH_CHECK: {
         readonly timeout: 5000;
-        readonly maxRetries: 2;
+        readonly maxRetries: 1;
         readonly retryDelay: 200;
     };
 };
@@ -140,7 +140,7 @@ export declare function getProviderRetryConfig(provider: string): {
     readonly maxDelay: 30000;
     readonly multiplier: 1.5;
 } | {
-    readonly maxAttempts: 2;
+    readonly maxAttempts: 1;
     readonly baseDelay: 200;
     readonly maxDelay: 5000;
     readonly multiplier: 1.5;

package/dist/constants/retry.d.ts

@@ -18,7 +18,7 @@ export declare const RETRY_ATTEMPTS: {
     /** Critical operations that must succeed */
     readonly CRITICAL: 5;
     /** Quick operations that should fail fast */
-    readonly QUICK: 2;
+    readonly QUICK: 1;
     /** Network operations prone to transient failures */
     readonly NETWORK: 4;
     /** Authentication operations */
@@ -122,7 +122,7 @@ export declare const PROVIDER_RETRY: {
     };
     /** Ollama retry configuration (local service) */
     readonly OLLAMA: {
-        readonly maxAttempts: 2;
+        readonly maxAttempts: 1;
         readonly baseDelay: 200;
         readonly maxDelay: 5000;
         readonly multiplier: 1.5;
@@ -141,7 +141,7 @@ export declare const OPERATION_RETRY: {
     };
     /** MCP operation retry config */
     readonly MCP_OPERATION: {
-        readonly maxAttempts: 2;
+        readonly maxAttempts: 1;
         readonly baseDelay: 200;
         readonly circuitBreaker: false;
     };
@@ -210,7 +210,7 @@ export declare const RetryUtils: {
         readonly maxDelay: 30000;
         readonly multiplier: 1.5;
     } | {
-        readonly maxAttempts: 2;
+        readonly maxAttempts: 1;
         readonly baseDelay: 200;
         readonly maxDelay: 5000;
         readonly multiplier: 1.5;

package/dist/constants/retry.js

@@ -18,7 +18,7 @@ export const RETRY_ATTEMPTS = {
     /** Critical operations that must succeed */
     CRITICAL: 5, // 5 attempts - High-importance operations
     /** Quick operations that should fail fast */
-    QUICK: 2, // 2 attempts - Fast operations, minimal retry
+    QUICK: 1, // 1 attempt, no retries — fail fast
     /** Network operations prone to transient failures */
     NETWORK: 4, // 4 attempts - Network operations
     /** Authentication operations */

package/dist/context/contextCompactor.d.ts

@@ -18,5 +18,5 @@ export declare class ContextCompactor {
     /**
      * Run the multi-stage compaction pipeline until messages fit within budget.
      */
-    compact(messages: ChatMessage[], targetTokens: number, memoryConfig?: Partial<ConversationMemoryConfig>): Promise<CompactionResult>;
+    compact(messages: ChatMessage[], targetTokens: number, memoryConfig?: Partial<ConversationMemoryConfig>, requestId?: string): Promise<CompactionResult>;
 }

package/dist/context/contextCompactor.js

@@ -9,6 +9,7 @@
  * Stage 4: Sliding Window Truncation (fallback -- no LLM call)
  */
 import { estimateMessagesTokens } from "../utils/tokenEstimation.js";
+import { logger } from "../utils/logger.js";
 import { pruneToolOutputs } from "./stages/toolOutputPruner.js";
 import { deduplicateFileReads } from "./stages/fileReadDeduplicator.js";
 import { summarizeMessages } from "./stages/structuredSummarizer.js";
@@ -35,14 +36,21 @@ export class ContextCompactor {
     /**
      * Run the multi-stage compaction pipeline until messages fit within budget.
      */
-    async compact(messages, targetTokens, memoryConfig) {
+    async compact(messages, targetTokens, memoryConfig, requestId) {
+        const compactionStartTime = Date.now();
         const provider = this.config.provider || undefined;
         const tokensBefore = estimateMessagesTokens(messages, provider);
         const stagesUsed = [];
         let currentMessages = [...messages];
+        logger.info("[Compaction] Starting", {
+            requestId,
+            estimatedTokens: tokensBefore,
+            budgetTokens: targetTokens,
+        });
         // Stage 1: Tool Output Pruning
         if (this.config.enablePrune &&
             estimateMessagesTokens(currentMessages, provider) > targetTokens) {
+            const stageTokensBefore = estimateMessagesTokens(currentMessages, provider);
             const pruneResult = pruneToolOutputs(currentMessages, {
                 protectTokens: this.config.pruneProtectTokens,
                 minimumSavings: this.config.pruneMinimumSavings,
@@ -53,19 +61,37 @@ export class ContextCompactor {
                 currentMessages = pruneResult.messages;
                 stagesUsed.push("prune");
             }
+            const stageTokensAfter = estimateMessagesTokens(currentMessages, provider);
+            logger.info("[Compaction] Stage 1 (prune)", {
+                requestId,
+                ran: pruneResult.pruned,
+                tokensBefore: stageTokensBefore,
+                tokensAfter: stageTokensAfter,
+                saved: stageTokensBefore - stageTokensAfter,
+            });
         }
         // Stage 2: File Read Deduplication
         if (this.config.enableDeduplicate &&
             estimateMessagesTokens(currentMessages, provider) > targetTokens) {
+            const stageTokensBefore = estimateMessagesTokens(currentMessages, provider);
             const dedupResult = deduplicateFileReads(currentMessages);
             if (dedupResult.deduplicated) {
                 currentMessages = dedupResult.messages;
                 stagesUsed.push("deduplicate");
             }
+            const stageTokensAfter = estimateMessagesTokens(currentMessages, provider);
+            logger.info("[Compaction] Stage 2 (deduplicate)", {
+                requestId,
+                ran: dedupResult.deduplicated,
+                tokensBefore: stageTokensBefore,
+                tokensAfter: stageTokensAfter,
+                saved: stageTokensBefore - stageTokensAfter,
+            });
         }
         // Stage 3: LLM Summarization
         if (this.config.enableSummarize &&
             estimateMessagesTokens(currentMessages, provider) > targetTokens) {
+            const stageTokensBefore = estimateMessagesTokens(currentMessages, provider);
             try {
                 const summarizeResult = await summarizeMessages(currentMessages, {
                     provider: this.config.summarizationProvider,
@@ -77,14 +103,30 @@ export class ContextCompactor {
                     currentMessages = summarizeResult.messages;
                     stagesUsed.push("summarize");
                 }
+                const stageTokensAfter = estimateMessagesTokens(currentMessages, provider);
+                logger.info("[Compaction] Stage 3 (summarize)", {
+                    requestId,
+                    ran: summarizeResult.summarized,
+                    tokensBefore: stageTokensBefore,
+                    tokensAfter: stageTokensAfter,
+                    saved: stageTokensBefore - stageTokensAfter,
+                });
             }
             catch {
+                logger.info("[Compaction] Stage 3 (summarize)", {
+                    requestId,
+                    ran: false,
+                    tokensBefore: stageTokensBefore,
+                    tokensAfter: stageTokensBefore,
+                    saved: 0,
+                });
                 // Summarization failed, fall through to truncation
             }
         }
         // Stage 4: Sliding Window Truncation (fallback)
         if (this.config.enableTruncate &&
             estimateMessagesTokens(currentMessages, provider) > targetTokens) {
+            const stageTokensBefore = estimateMessagesTokens(currentMessages, provider);
             const truncResult = truncateWithSlidingWindow(currentMessages, {
                 fraction: this.config.truncationFraction,
             });
@@ -92,8 +134,24 @@ export class ContextCompactor {
                 currentMessages = truncResult.messages;
                 stagesUsed.push("truncate");
             }
+            const stageTokensAfter = estimateMessagesTokens(currentMessages, provider);
+            logger.info("[Compaction] Stage 4 (truncate)", {
+                requestId,
+                ran: truncResult.truncated,
+                tokensBefore: stageTokensBefore,
+                tokensAfter: stageTokensAfter,
+                saved: stageTokensBefore - stageTokensAfter,
+            });
         }
         const tokensAfter = estimateMessagesTokens(currentMessages, provider);
+        logger.info("[Compaction] Complete", {
+            requestId,
+            tokensBefore,
+            tokensAfter,
+            totalSaved: tokensBefore - tokensAfter,
+            stagesUsed,
+            durationMs: Date.now() - compactionStartTime,
+        });
         return {
             compacted: stagesUsed.length > 0,
             stagesUsed,

package/dist/context/summarizationEngine.d.ts

@@ -18,7 +18,7 @@ export declare class SummarizationEngine {
      * @param logPrefix - Prefix for log messages
      * @returns True if summarization was performed
      */
-    checkAndSummarize(session: SessionMemory, threshold: number, config: Partial<ConversationMemoryConfig>, logPrefix?: string): Promise<boolean>;
+    checkAndSummarize(session: SessionMemory, threshold: number, config: Partial<ConversationMemoryConfig>, logPrefix?: string, requestId?: string): Promise<boolean>;
     /**
      * Perform token-based summarization on a session.
      * Uses pointer-based, non-destructive approach.
@@ -27,7 +27,7 @@ export declare class SummarizationEngine {
      * @param config - Conversation memory configuration (partial allowed)
      * @param logPrefix - Prefix for log messages
      */
-    summarizeSession(session: SessionMemory, threshold: number, config: Partial<ConversationMemoryConfig>, logPrefix?: string): Promise<void>;
+    summarizeSession(session: SessionMemory, threshold: number, config: Partial<ConversationMemoryConfig>, logPrefix?: string, requestId?: string): Promise<void>;
     /**
      * Estimate total tokens for a message array.
      * @param messages - Array of chat messages