@prestyj/ai 5.27.0 → 5.28.1

package/README.md

@@ -40,10 +40,10 @@ Tool parameters are Zod schemas. Converted to JSON Schema at the provider bounda
 
 | Provider | Models | Notes |
 |---|---|---|
-| `anthropic` | Claude Fable 5.1, Opus 5.5, Opus 5, Sonnet 5, Haiku 4.5 | Extended thinking, prompt caching, server-side compaction |
+| `anthropic` | Claude Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Sonnet 5, Haiku 4.5 | Extended thinking, prompt caching, server-side compaction |
 | `openai` | GPT-4.1, o3, o4-mini | Supports OAuth (codex endpoint) and API keys |
 | `glm` | GLM-5.3 | Z.AI platform, OpenAI-compatible |
-| `moonshot` | Kimi K3, Kimi K2.7 Code | Moonshot platform, OpenAI-compatible |
+| `moonshot` | Kimi K3, Kimi K2.8 Preview, Kimi K2.7 Code | Moonshot platform, OpenAI-compatible |
 
 ---
 

package/dist/index.cjs

@@ -1100,7 +1100,7 @@ function isAdaptiveThinkingModel(model) {
 function toAnthropicThinking(level, maxTokens, model) {
   if (isAdaptiveThinkingModel(model)) {
     let effort = level;
-    if (effort === "xhigh" && !/opus-5|opus-4-8|opus-4-7/.test(model)) {
+    if (effort === "xhigh" && !/opus-5|opus-4[-.]8|opus-4[-.]7|sonnet-5[-.]5/.test(model)) {
       effort = "high";
     }
     return {
@@ -2103,7 +2103,7 @@ async function* runStream2(options) {
   const endpointKey = reasoningFieldKey(providerName, options.baseUrl, options.model);
   const client = createClient2(options);
   const isLocal = options.provider === "local";
-  const isKimiK3 = options.provider === "moonshot" && options.model === "kimi-k3";
+  const isKimiK3 = options.provider === "moonshot" && (options.model === "kimi-k3" || options.model === "kimi-for-coding");
   const isManagedKimiK3 = isKimiK3 && options.baseUrl?.replace(/\/+$/, "").endsWith("/coding/v1") === true;
   const k3Effort = options.thinking ? toKimiK3Effort(options.thinking) : void 0;
   const isKimiK27 = options.provider === "moonshot" && options.model.startsWith("kimi-k2.7-code");
@@ -2577,7 +2577,7 @@ function extractRequestIdFromMessage(message) {
 
 // src/providers/openai-codex.ts
 var DEFAULT_BASE_URL = "https://chatgpt.com/backend-api";
-var CODEX_CLIENT_VERSION = "0.155.1";
+var CODEX_CLIENT_VERSION = "0.159.1";
 var CODEX_REQUEST_COMPRESSION_MIN_BYTES = 16 * 1024;
 var zstdInitPromise;
 async function encodeCodexRequest(body) {
@@ -2623,7 +2623,7 @@ async function encodeCodexRequest(body) {
   }
 }
 function usesResponsesLite(model) {
-  return model.startsWith("gpt-5.6-") || model.startsWith("gpt-6-");
+  return model.startsWith("gpt-5.6-") || model.startsWith("gpt-6-") || model.startsWith("gpt-6.");
 }
 function outputTextKey(itemId, contentIndex) {
   return `${itemId ?? ""}:${contentIndex ?? 0}`;
@@ -2745,7 +2745,7 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
     if (response.status === 400 && message === `The '${options.model}' model is not supported when using Codex with a ChatGPT account.`) {
       hint = "This model is not available through your ChatGPT account. Switch to a model listed for OpenAI via the model selector, or check your ChatGPT usage limits.";
     } else if (response.status === 404 && text.includes("does not exist")) {
-      hint = "This model is not in OpenAI's current catalog for your ChatGPT account. Switch to GPT-6 Astra, GPT-6 Sol, or GPT-6 Luna via the model selector.";
+      hint = "This model is not in OpenAI's current catalog for your ChatGPT account. Switch to GPT-6 Astra, GPT-6.1 Sol, or GPT-6 Luna via the model selector.";
     }
     throw new ProviderError("openai", message, {
       statusCode: response.status,
@@ -2759,6 +2759,8 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
   const contentParts = [];
   let textAccum = "";
   const toolCalls = /* @__PURE__ */ new Map();
+  const finishedToolCalls = /* @__PURE__ */ new Set();
+  let terminal;
   const orderedItems = [];
   const outputItemTypes = /* @__PURE__ */ new Map();
   const outputTextByPart = /* @__PURE__ */ new Map();
@@ -2902,6 +2904,7 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
       for (const [key, tc] of toolCalls) {
         if (key.endsWith(`|${itemId}`)) {
           tc.argsJson = argsStr;
+          finishedToolCalls.add(key);
           break;
         }
       }
@@ -2927,6 +2930,7 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
         const id = `${callId}|${itemId}`;
         const tc = toolCalls.get(id);
         if (tc) {
+          finishedToolCalls.add(id);
           orderedItems.push({ kind: "tool", id });
           const args = parseToolArguments(tc.argsJson);
           yield {
@@ -2938,8 +2942,17 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
         }
       }
     }
-    if (type === "response.completed" || type === "response.done") {
+    if (type === "response.completed" || type === "response.done" || type === "response.incomplete") {
       const resp = event.response;
+      if (type === "response.incomplete" || resp?.status === "incomplete") {
+        const details = resp?.incomplete_details;
+        terminal = {
+          status: "incomplete",
+          reason: typeof details?.reason === "string" ? details.reason : void 0
+        };
+      } else {
+        terminal = { status: "completed" };
+      }
       const usage = resp?.usage;
       if (usage) {
         cacheRead = usage.input_tokens_details?.cached_tokens ?? 0;
@@ -2949,6 +2962,25 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
       }
     }
   }
+  if (!terminal) {
+    throw new ProviderError("openai", "Stream ended before completion (no response.completed).", {
+      statusCode: 504
+    });
+  }
+  const droppedToolCalls = [...toolCalls.keys()].filter((id) => !finishedToolCalls.has(id)).length;
+  if (terminal.status === "completed") {
+    for (const [id, tc] of toolCalls) {
+      if (!finishedToolCalls.has(id)) {
+        throw new ProviderError(
+          "openai",
+          `Codex reply completed with an unfinished tool call: ${tc.name} (${id}).`,
+          { statusCode: 502 }
+        );
+      }
+    }
+  } else {
+    providerDiag("codex_incomplete", { reason: terminal.reason ?? null, droppedToolCalls });
+  }
   const seenTool = /* @__PURE__ */ new Set();
   let textInserted = false;
   for (const entry of orderedItems) {
@@ -2975,7 +3007,7 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
     contentParts.push({ type: "text", text: textAccum });
   }
   for (const [id, tc] of toolCalls) {
-    if (seenTool.has(id)) continue;
+    if (seenTool.has(id) || !finishedToolCalls.has(id)) continue;
     seenTool.add(id);
     contentParts.push({
       type: "tool_call",
@@ -2984,8 +3016,15 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
       args: parseToolArguments(tc.argsJson)
     });
   }
+  if (droppedToolCalls > 0) {
+    let last = contentParts.at(-1);
+    while (last?.type === "raw" && isEncryptedReasoning(last.data)) {
+      contentParts.pop();
+      last = contentParts.at(-1);
+    }
+  }
   const hasToolCalls = contentParts.some((p) => p.type === "tool_call");
-  const stopReason = hasToolCalls ? "tool_use" : "end_turn";
+  const stopReason = terminal.status === "incomplete" ? incompleteStopReason(terminal.reason) : hasToolCalls ? "tool_use" : "end_turn";
   const streamResponse = {
     message: {
       role: "assistant",
@@ -3002,6 +3041,11 @@ async function* runStream3(options, retriedWithoutReasoning = false) {
   yield { type: "done", stopReason };
   return streamResponse;
 }
+function incompleteStopReason(reason) {
+  if (reason === "max_output_tokens") return "max_tokens";
+  if (reason === "content_filter") return "refusal";
+  return "error";
+}
 async function* parseSSE(body) {
   for await (const event of readSseStream(body)) {
     const data = event.data.trim();
@@ -4137,6 +4181,8 @@ var CIRCULAR = "[CIRCULAR]";
 var SENSITIVE_NAME = /(?:^|[_-])(?:api[_-]?key|access[_-]?token|refresh[_-]?token|token|key|auth(?:orization)?|bearer|cookie|credential|private[_-]?key|password|passwd|secret)(?:$|[_-])/i;
 var ENV_SECRET_ASSIGNMENT = /\b((?:[A-Z0-9]+_)*(?:API_?KEY|ACCESS_TOKEN|REFRESH_TOKEN|TOKEN|KEY|AUTH|AUTHORIZATION|BEARER|CREDENTIALS?|PASSWORD|PASSWD|SECRET))\b(\s*[=:]\s*)(["']?)(?!\$|process\.env|os\.environ|import\.meta|env\.)(?=[^\s,"';}]*\d)([^\s,"';}=$][^\s,"';}]{7,})\3/g;
 var COMPACT_SECRET_ASSIGNMENT = /\b((?:[a-z0-9]+[_-])*(?:api[_-]?key|access[_-]?token|refresh[_-]?token|token|auth|authorization|credentials?|password|passwd|secret)|(?:[a-z0-9]+[_-])+key)=(["']?)(?!\$)([^\s,"'&;}=][^\s,"'&;}]{7,})\2/gi;
+var URL_USERINFO = /\b([a-z][a-z0-9+.-]*:\/\/)[^\s/?#@:"'`<>]*:[^\s/?#"'`<>]+@/gi;
+var URL_USERINFO_TO_FIRST_AT = /\b([a-z][a-z0-9+.-]*:\/\/)[^\s/@:]*:[^\s/@]+@/gi;
 function escaped(value) {
   return value.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
 }
@@ -4160,7 +4206,8 @@ function redactText(text, options = {}) {
     /-----BEGIN (?:[A-Z0-9 ]+ )?PRIVATE KEY-----[\s\S]*?-----END (?:[A-Z0-9 ]+ )?PRIVATE KEY-----/g,
     REDACTED
   );
-  result = result.replace(/\b([a-z][a-z0-9+.-]*:\/\/)[^\s/@:]+:[^\s/@]+@/gi, `$1${REDACTED}@`);
+  result = result.replace(URL_USERINFO, `$1${REDACTED}@`);
+  result = result.replace(URL_USERINFO_TO_FIRST_AT, `$1${REDACTED}@`);
   result = result.replace(
     /\b(authorization\s*[:=]\s*)(?:bearer|basic)\s+[^\s,;]+/gi,
     `$1${REDACTED}`
@@ -4204,7 +4251,7 @@ function isMediaObject(value) {
 function redactValue(value, options = {}) {
   const maxDepth = options.maxDepth ?? 20;
   const maxEntries = options.maxEntries ?? 1e4;
-  const seen = /* @__PURE__ */ new WeakSet();
+  const ancestors = /* @__PURE__ */ new WeakSet();
   let entries = 0;
   const visit = (current, depth, sensitive = false) => {
     if (typeof current === "string") {
@@ -4218,8 +4265,15 @@ function redactValue(value, options = {}) {
     if (isBinary(current)) return current;
     if (current instanceof Date) return new Date(current.getTime());
     if (depth >= maxDepth) return TRUNCATED;
-    if (seen.has(current)) return CIRCULAR;
-    seen.add(current);
+    if (ancestors.has(current)) return CIRCULAR;
+    ancestors.add(current);
+    try {
+      return cloneObject(current, depth);
+    } finally {
+      ancestors.delete(current);
+    }
+  };
+  const cloneObject = (current, depth) => {
     if (current instanceof Error) {
       const error = {
         name: current.name,