@kenkaiiii/gg-ai 5.68.0 → 5.68.1

package/dist/index.d.cts

@@ -186,8 +186,24 @@ interface Usage {
         webFetchRequests?: number;
     };
 }
+interface PreparedContext {
+    /** Read-only observation after generic sanitization/image limiting, before provider encoding.
+     * Do not log or retain these messages: they may contain secrets and image data. */
+    messages: readonly Message[];
+    tools: readonly {
+        name: string;
+        description: string;
+        parameters: Record<string, unknown>;
+    }[];
+    imagesBefore: number;
+    imagesAfter: number;
+    /** First message whose images were removed for this request, or null. */
+    firstImageDropMessage: number | null;
+}
 interface StreamOptions {
     provider: Provider;
+    /** Optional, synchronous diagnostics observer. Exceptions cannot fail a model request. */
+    onContextPrepared?: (context: PreparedContext) => void;
     model: string;
     messages: Message[];
     tools?: Tool[];
@@ -582,7 +598,8 @@ type JsonSchema = Record<string, unknown>;
 declare function resolveToolSchema(tool: Tool): JsonSchema;
 
 /**
- * Cap historical images before provider dispatch, removing the oldest first.
+ * Cap historical images before provider dispatch, removing the oldest first, in batches so the
+ * cached conversation prefix stays byte-identical between requests.
  * The persisted/live conversation is never mutated; only modified messages and
  * tool results are cloned for the outgoing request.
  */
@@ -693,4 +710,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, environmentSecrets, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreparedContext, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, environmentSecrets, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };

package/dist/index.d.ts

@@ -186,8 +186,24 @@ interface Usage {
         webFetchRequests?: number;
     };
 }
+interface PreparedContext {
+    /** Read-only observation after generic sanitization/image limiting, before provider encoding.
+     * Do not log or retain these messages: they may contain secrets and image data. */
+    messages: readonly Message[];
+    tools: readonly {
+        name: string;
+        description: string;
+        parameters: Record<string, unknown>;
+    }[];
+    imagesBefore: number;
+    imagesAfter: number;
+    /** First message whose images were removed for this request, or null. */
+    firstImageDropMessage: number | null;
+}
 interface StreamOptions {
     provider: Provider;
+    /** Optional, synchronous diagnostics observer. Exceptions cannot fail a model request. */
+    onContextPrepared?: (context: PreparedContext) => void;
     model: string;
     messages: Message[];
     tools?: Tool[];
@@ -582,7 +598,8 @@ type JsonSchema = Record<string, unknown>;
 declare function resolveToolSchema(tool: Tool): JsonSchema;
 
 /**
- * Cap historical images before provider dispatch, removing the oldest first.
+ * Cap historical images before provider dispatch, removing the oldest first, in batches so the
+ * cached conversation prefix stays byte-identical between requests.
  * The persisted/live conversation is never mutated; only modified messages and
  * tool results are cloned for the outgoing request.
  */
@@ -693,4 +710,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, environmentSecrets, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, GGAIError, type ImageContent, type Message, type MessageProvenance, type MessageProvenanceKind, type MessageProvenanceSource, type MessageProvenanceVisibility, type PalsuModelConfig, type PalsuModelHandle, type PalsuProviderConfig, type PalsuProviderHandle, type PalsuProviderState, type PalsuResponse, type PalsuResponseFactory, type PreparedContext, type Provider, type ProviderDiagnosticFn, type ProviderEntry, ProviderError, type ProviderStreamFn, REDACTED as REDACTION_MARKER, type RawContent, type RedactionOptions, type ServerToolCall, type ServerToolCallEvent, type ServerToolDefinition, type ServerToolResult, type ServerToolResultEvent, type StopReason, type StreamEvent, type StreamOptions, type StreamResponse, StreamResult, type SystemMessage, type TextContent, type TextDeltaEvent, type ThinkingContent, type ThinkingDeltaEvent, type ThinkingLevel, type Tool, type ToolCall, type ToolCallDeltaEvent, type ToolCallDoneEvent, type ToolChoice, type ToolResult, type ToolResultContent, type ToolResultMessage, type Usage, type UserMessage, type VideoContent, clampProviderContextImages, classifyProviderError, environmentSecrets, formatError, formatErrorForDisplay, hasLoneSurrogate, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, resolveToolSchema, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };

package/dist/index.js

@@ -743,10 +743,17 @@ function countContextImages(messages) {
   }
   return count;
 }
+var IMAGE_DROP_BATCH = 30;
+function providerImageDropCount(imageCount2, budget) {
+  const overflow = imageCount2 - budget;
+  if (overflow <= 0) return 0;
+  const batch = Math.max(1, Math.min(IMAGE_DROP_BATCH, Math.floor(budget / 3)));
+  return Math.min(imageCount2, Math.ceil(overflow / batch) * batch);
+}
 function clampProviderContextImages(messages, provider, supportsImages) {
   if (supportsImages === false) return messages;
   const budget = PROVIDER_IMAGE_BUDGETS[provider] ?? 5;
-  let remainingToRemove = countContextImages(messages) - budget;
+  let remainingToRemove = providerImageDropCount(countContextImages(messages), budget);
   if (remainingToRemove <= 0) return messages;
   return messages.map((message) => {
     if (message.role === "user" && Array.isArray(message.content)) {
@@ -3841,6 +3848,41 @@ function sanitizeMessagesForWire(messages) {
   return sanitized ?? messages;
 }
 
+// src/utils/context-observation.ts
+function imageCount(message) {
+  if (!message || !Array.isArray(message.content)) return 0;
+  if (message.role === "user")
+    return message.content.filter((part) => part.type === "image").length;
+  if (message.role !== "tool") return 0;
+  return message.content.reduce(
+    (count, result) => count + (Array.isArray(result.content) ? result.content.filter((part) => part.type === "image").length : 0),
+    0
+  );
+}
+function observePreparedContext(before, after, tools) {
+  let imagesBefore = 0;
+  let imagesAfter = 0;
+  let firstImageDropMessage = null;
+  for (let index = 0; index < before.length; index++) {
+    const oldCount = imageCount(before[index]);
+    const newCount = imageCount(after[index]);
+    imagesBefore += oldCount;
+    imagesAfter += newCount;
+    if (firstImageDropMessage === null && newCount < oldCount) firstImageDropMessage = index;
+  }
+  return {
+    messages: after,
+    tools: tools.map((tool) => ({
+      name: tool.name,
+      description: tool.description,
+      parameters: resolveToolSchema(tool)
+    })),
+    imagesBefore,
+    imagesAfter,
+    firstImageDropMessage
+  };
+}
+
 // src/stream.ts
 var GLM_CODING_BASE_URL = "https://api.z.ai/api/coding/paas/v4";
 var KIMI_CODE_USER_AGENT = `kimi-code-cli/${process.env.KIMI_CODE_VERSION ?? "1.0.11"}`;
@@ -3988,6 +4030,14 @@ function stream(options) {
     options.provider,
     options.supportsImages
   );
+  if (options.onContextPrepared) {
+    try {
+      options.onContextPrepared(
+        observePreparedContext(wireMessages, messages, options.tools ?? [])
+      );
+    } catch {
+    }
+  }
   return entry.stream(messages === options.messages ? options : { ...options, messages });
 }
 function stripMessageProvenance(messages) {