@prestyj/ai 5.10.0 → 5.11.0

package/dist/index.d.cts

@@ -77,19 +77,31 @@ interface RawContent {
     data: Record<string, unknown>;
 }
 type ContentPart = TextContent | ThinkingContent | ImageContent | VideoContent | ToolCall | ServerToolCall | ServerToolResult | RawContent;
-interface SystemMessage {
+type MessageProvenanceSource = "human" | "agent" | "runtime";
+type MessageProvenanceKind = "prompt" | "steering" | "notification" | "completion_gate" | "review_follow_up" | "continuation" | "model_switch" | "automation" | "compaction_summary" | "compaction_ack";
+type MessageProvenanceVisibility = "transcript" | "hidden" | "summary";
+/** Internal message metadata. `stream()` removes it before provider dispatch. */
+interface MessageProvenance {
+    source: MessageProvenanceSource;
+    kind: MessageProvenanceKind;
+    visibility: MessageProvenanceVisibility;
+}
+interface MessageMetadata {
+    provenance?: MessageProvenance;
+}
+interface SystemMessage extends MessageMetadata {
     role: "system";
     content: string;
 }
-interface UserMessage {
+interface UserMessage extends MessageMetadata {
     role: "user";
     content: string | (TextContent | ImageContent | VideoContent)[];
 }
-interface AssistantMessage {
+interface AssistantMessage extends MessageMetadata {
     role: "assistant";
     content: string | ContentPart[];
 }
-interface ToolResultMessage {
+interface ToolResultMessage extends MessageMetadata {
     role: "tool";
     content: ToolResult[];
 }
@@ -161,7 +173,10 @@ interface StreamResponse {
 }
 interface Usage {
     inputTokens: number;
+    /** Total billed output tokens, including reasoning tokens when the provider reports them separately. */
     outputTokens: number;
+    /** Reasoning/thinking-token subset of outputTokens. */
+    reasoningTokens?: number;
     cacheRead?: number;
     cacheWrite?: number;
     serverToolUse?: {
@@ -490,6 +505,21 @@ declare function redactText(text: string, options?: RedactionOptions): string;
  */
 declare function redactValue<T>(value: T, options?: RedactionOptions): T;
 
+/** True when the string contains at least one unpaired surrogate. */
+declare function hasLoneSurrogate(text: string): boolean;
+/** Replace unpaired surrogates with U+FFFD; returns the input when already valid. */
+declare function toWellFormedText(text: string): string;
+/** `text.slice(0, chars)` that never cuts an astral character in half. */
+declare function sliceHead(text: string, chars: number): string;
+/** `text.slice(-chars)` that never cuts an astral character in half. */
+declare function sliceTail(text: string, chars: number): string;
+/**
+ * Strip unpaired surrogates from everything headed for the wire. Returns the
+ * same array (and same message objects) when the history is already valid, so
+ * the clean path stays allocation-free.
+ */
+declare function sanitizeMessagesForWire(messages: Message[]): Message[];
+
 /**
  * Provider-level diagnostic hook. Mirrors the pattern used by gg-agent's
  * setStreamDiagnostic — the host app wires a callback (typically writing to
@@ -612,4 +642,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, EZCoderAIError, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, type ImageContent, type Message, 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, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, setProviderDiagnostic, stream, toAnthropicMessages, toOpenAIMessages };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, EZCoderAIError, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, 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, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };

package/dist/index.d.ts

@@ -77,19 +77,31 @@ interface RawContent {
     data: Record<string, unknown>;
 }
 type ContentPart = TextContent | ThinkingContent | ImageContent | VideoContent | ToolCall | ServerToolCall | ServerToolResult | RawContent;
-interface SystemMessage {
+type MessageProvenanceSource = "human" | "agent" | "runtime";
+type MessageProvenanceKind = "prompt" | "steering" | "notification" | "completion_gate" | "review_follow_up" | "continuation" | "model_switch" | "automation" | "compaction_summary" | "compaction_ack";
+type MessageProvenanceVisibility = "transcript" | "hidden" | "summary";
+/** Internal message metadata. `stream()` removes it before provider dispatch. */
+interface MessageProvenance {
+    source: MessageProvenanceSource;
+    kind: MessageProvenanceKind;
+    visibility: MessageProvenanceVisibility;
+}
+interface MessageMetadata {
+    provenance?: MessageProvenance;
+}
+interface SystemMessage extends MessageMetadata {
     role: "system";
     content: string;
 }
-interface UserMessage {
+interface UserMessage extends MessageMetadata {
     role: "user";
     content: string | (TextContent | ImageContent | VideoContent)[];
 }
-interface AssistantMessage {
+interface AssistantMessage extends MessageMetadata {
     role: "assistant";
     content: string | ContentPart[];
 }
-interface ToolResultMessage {
+interface ToolResultMessage extends MessageMetadata {
     role: "tool";
     content: ToolResult[];
 }
@@ -161,7 +173,10 @@ interface StreamResponse {
 }
 interface Usage {
     inputTokens: number;
+    /** Total billed output tokens, including reasoning tokens when the provider reports them separately. */
     outputTokens: number;
+    /** Reasoning/thinking-token subset of outputTokens. */
+    reasoningTokens?: number;
     cacheRead?: number;
     cacheWrite?: number;
     serverToolUse?: {
@@ -490,6 +505,21 @@ declare function redactText(text: string, options?: RedactionOptions): string;
  */
 declare function redactValue<T>(value: T, options?: RedactionOptions): T;
 
+/** True when the string contains at least one unpaired surrogate. */
+declare function hasLoneSurrogate(text: string): boolean;
+/** Replace unpaired surrogates with U+FFFD; returns the input when already valid. */
+declare function toWellFormedText(text: string): string;
+/** `text.slice(0, chars)` that never cuts an astral character in half. */
+declare function sliceHead(text: string, chars: number): string;
+/** `text.slice(-chars)` that never cuts an astral character in half. */
+declare function sliceTail(text: string, chars: number): string;
+/**
+ * Strip unpaired surrogates from everything headed for the wire. Returns the
+ * same array (and same message objects) when the history is already valid, so
+ * the clean path stays allocation-free.
+ */
+declare function sanitizeMessagesForWire(messages: Message[]): Message[];
+
 /**
  * Provider-level diagnostic hook. Mirrors the pattern used by gg-agent's
  * setStreamDiagnostic — the host app wires a callback (typically writing to
@@ -612,4 +642,4 @@ interface PalsuProviderConfig {
  */
 declare function registerPalsuProvider(config?: PalsuProviderConfig): PalsuProviderHandle;
 
-export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, EZCoderAIError, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, type ImageContent, type Message, 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, isHardBillingMessage, isUsageLimitError, localWireModelId, palsuAssistantMessage, palsuText, palsuThinking, palsuToolCall, prewarmAnthropicCache, providerRegistry, redactText, redactValue, registerPalsuProvider, setProviderDiagnostic, stream, toAnthropicMessages, toOpenAIMessages };
+export { type AssistantMessage, type CacheRetention, type ContentPart, type DoneEvent, EZCoderAIError, type ErrorEvent, type ErrorSource, EventStream, type FormattedError, 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, sanitizeMessagesForWire, setProviderDiagnostic, sliceHead, sliceTail, stream, toAnthropicMessages, toOpenAIMessages, toWellFormedText };

package/dist/index.js

@@ -3319,14 +3319,16 @@ async function* runStream4(options) {
   let thinkingAccum = "";
   let stopReason = "end_turn";
   let inputTokens = 0;
-  let outputTokens = 0;
+  let candidateTokens = 0;
+  let reasoningTokens = 0;
   let cacheRead = 0;
   let toolIndex = 0;
   const handleResponse = function* (chunk) {
     const usage = usageFromResponse(chunk);
     if (usage) {
       inputTokens = usage.promptTokenCount ?? inputTokens;
-      outputTokens = usage.candidatesTokenCount ?? outputTokens;
+      candidateTokens = usage.candidatesTokenCount ?? candidateTokens;
+      reasoningTokens = usage.thoughtsTokenCount ?? reasoningTokens;
       cacheRead = usage.cachedContentTokenCount ?? cacheRead;
     }
     const reason = finishReasonFromResponse(chunk);
@@ -3382,6 +3384,7 @@ async function* runStream4(options) {
   }
   if (pendingToolCalls.length > 0) stopReason = "tool_use";
   const adjustedInputTokens = Math.max(0, inputTokens - cacheRead);
+  const outputTokens = candidateTokens + reasoningTokens;
   const streamResponse = {
     message: {
       role: "assistant",
@@ -3391,6 +3394,7 @@ async function* runStream4(options) {
     usage: {
       inputTokens: adjustedInputTokens,
       outputTokens,
+      ...reasoningTokens > 0 ? { reasoningTokens } : {},
       ...cacheRead > 0 ? { cacheRead } : {}
     }
   };
@@ -3439,6 +3443,140 @@ var ProviderRegistryImpl = class {
 };
 var providerRegistry = new ProviderRegistryImpl();
 
+// src/utils/well-formed.ts
+var LONE_SURROGATE = /[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/;
+var LONE_SURROGATE_GLOBAL = new RegExp(LONE_SURROGATE, "g");
+var REPLACEMENT = "\uFFFD";
+function hasLoneSurrogate(text) {
+  const isWellFormed = text.isWellFormed;
+  if (typeof isWellFormed === "function") return !isWellFormed.call(text);
+  return LONE_SURROGATE.test(text);
+}
+function toWellFormedText(text) {
+  if (!hasLoneSurrogate(text)) return text;
+  const toWellFormed = text.toWellFormed;
+  if (typeof toWellFormed === "function") return toWellFormed.call(text);
+  return text.replace(LONE_SURROGATE_GLOBAL, REPLACEMENT);
+}
+function isHighSurrogate(code) {
+  return code !== void 0 && code >= 55296 && code <= 56319;
+}
+function isLowSurrogate(code) {
+  return code !== void 0 && code >= 56320 && code <= 57343;
+}
+function sliceHead(text, chars) {
+  if (chars <= 0) return "";
+  if (chars >= text.length) return text;
+  const end = isHighSurrogate(text.charCodeAt(chars - 1)) ? chars - 1 : chars;
+  return text.slice(0, end);
+}
+function sliceTail(text, chars) {
+  if (chars <= 0) return "";
+  if (chars >= text.length) return text;
+  const start = text.length - chars;
+  return text.slice(isLowSurrogate(text.charCodeAt(start)) ? start + 1 : start);
+}
+function sanitizeJsonValue(value) {
+  if (typeof value === "string") return toWellFormedText(value);
+  if (Array.isArray(value)) {
+    let changed = false;
+    const next = value.map((item) => {
+      const sanitized = sanitizeJsonValue(item);
+      if (sanitized !== item) changed = true;
+      return sanitized;
+    });
+    return changed ? next : value;
+  }
+  if (value !== null && typeof value === "object") {
+    let changed = false;
+    const next = {};
+    for (const [key, item] of Object.entries(value)) {
+      const sanitizedKey = toWellFormedText(key);
+      const sanitized = sanitizeJsonValue(item);
+      if (sanitizedKey !== key || sanitized !== item) changed = true;
+      next[sanitizedKey] = sanitized;
+    }
+    return changed ? next : value;
+  }
+  return value;
+}
+function sanitizeRecord(value) {
+  return sanitizeJsonValue(value);
+}
+function sanitizePart(part) {
+  switch (part.type) {
+    case "text":
+    case "thinking": {
+      const text = toWellFormedText(part.text);
+      return text === part.text ? part : { ...part, text };
+    }
+    case "tool_call": {
+      const args = sanitizeRecord(part.args);
+      return args === part.args ? part : { ...part, args };
+    }
+    case "server_tool_call": {
+      const input = sanitizeJsonValue(part.input);
+      return input === part.input ? part : { ...part, input };
+    }
+    case "server_tool_result": {
+      const data = sanitizeJsonValue(part.data);
+      return data === part.data ? part : { ...part, data };
+    }
+    case "raw": {
+      const data = sanitizeRecord(part.data);
+      return data === part.data ? part : { ...part, data };
+    }
+    default:
+      return part;
+  }
+}
+function sanitizeParts(parts) {
+  let changed = false;
+  const next = parts.map((part) => {
+    const sanitized = sanitizePart(part);
+    if (sanitized !== part) changed = true;
+    return sanitized;
+  });
+  return changed ? next : parts;
+}
+function sanitizeToolResultContent(content) {
+  if (typeof content === "string") return toWellFormedText(content);
+  return sanitizeParts(content);
+}
+function sanitizeToolResults(results) {
+  let changed = false;
+  const next = results.map((result) => {
+    const content = sanitizeToolResultContent(result.content);
+    if (content === result.content) return result;
+    changed = true;
+    return { ...result, content };
+  });
+  return changed ? next : results;
+}
+function sanitizeMessage(message) {
+  if (message.role === "tool") {
+    const content2 = sanitizeToolResults(message.content);
+    return content2 === message.content ? message : { ...message, content: content2 };
+  }
+  if (typeof message.content === "string") {
+    const content2 = toWellFormedText(message.content);
+    return content2 === message.content ? message : { ...message, content: content2 };
+  }
+  const content = sanitizeParts(message.content);
+  return content === message.content ? message : { ...message, content };
+}
+function sanitizeMessagesForWire(messages) {
+  let sanitized;
+  for (let index = 0; index < messages.length; index++) {
+    const message = messages[index];
+    const next = sanitizeMessage(message);
+    if (next === message) continue;
+    sanitized ??= messages.slice();
+    sanitized[index] = next;
+  }
+  return sanitized ?? messages;
+}
+
 // 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"}`;
@@ -3560,13 +3698,25 @@ function stream(options) {
   if (options.supportsVideo !== true && messagesContainVideo(options.messages)) {
     throw new VideoUnsupportedError();
   }
+  const wireMessages = stripMessageProvenance(options.messages);
   const messages = clampProviderContextImages(
-    options.messages,
+    sanitizeMessagesForWire(wireMessages),
     options.provider,
     options.supportsImages
   );
   return entry.stream(messages === options.messages ? options : { ...options, messages });
 }
+function stripMessageProvenance(messages) {
+  let stripped;
+  for (let index = 0; index < messages.length; index++) {
+    const message = messages[index];
+    if (!message.provenance) continue;
+    stripped ??= messages.slice();
+    const { provenance: _provenance, ...wireMessage } = message;
+    stripped[index] = wireMessage;
+  }
+  return stripped ?? messages;
+}
 function messagesContainVideo(messages) {
   for (const msg of messages) {
     if (typeof msg.content === "string" || !Array.isArray(msg.content)) continue;
@@ -3951,6 +4101,7 @@ export {
   environmentSecrets,
   formatError,
   formatErrorForDisplay,
+  hasLoneSurrogate,
   isHardBillingMessage,
   isUsageLimitError,
   localWireModelId,
@@ -3963,9 +4114,13 @@ export {
   redactText,
   redactValue,
   registerPalsuProvider,
+  sanitizeMessagesForWire,
   setProviderDiagnostic,
+  sliceHead,
+  sliceTail,
   stream,
   toAnthropicMessages,
-  toOpenAIMessages
+  toOpenAIMessages,
+  toWellFormedText
 };
 //# sourceMappingURL=index.js.map