@prestyj/ai 5.3.0 → 5.4.0

package/dist/index.cjs

@@ -172,7 +172,7 @@ function formatError(err) {
         provider: err.provider,
         statusCode: err.statusCode,
         ...err.requestId ? { requestId: err.requestId } : {},
-        guidance: "Request access via your Anthropic account team (see platform.claude.com/docs/en/about-claude/models/overview), or switch to claude-fable-5 with /model \u2014 same underlying model, generally available."
+        guidance: "Request access via your Anthropic account team (see platform.claude.com/docs/en/about-claude/models/overview), or switch to Claude Fable 5 via the model selector \u2014 same underlying model, generally available."
       };
     }
     if (isUsageLimitError(err)) {
@@ -214,7 +214,7 @@ function finaliseBySource(source, message, requestId, hint) {
         headline: "Network error \u2014 couldn't reach the provider.",
         source,
         message,
-        guidance: hint ?? "Check your internet connection. Not a ezcoder issue \u2014 retry shortly.",
+        guidance: hint ?? "Check your internet connection. Not a EZ Coder issue \u2014 retry shortly.",
         ...requestId ? { requestId } : {}
       };
     case "auth":
@@ -222,7 +222,7 @@ function finaliseBySource(source, message, requestId, hint) {
         headline: "Authentication issue.",
         source,
         message,
-        guidance: hint ?? "Run `ezcoder login` to refresh your credentials.",
+        guidance: hint ?? "Re-authenticate to refresh your credentials.",
         ...requestId ? { requestId } : {}
       };
     case "provider":
@@ -238,15 +238,15 @@ function finaliseBySource(source, message, requestId, hint) {
         headline: message,
         source,
         message: "",
-        guidance: hint ?? "Only Kimi, Gemini, MiniMax, and MiMo-V2.5 can analyze video. Switch with /model.",
+        guidance: hint ?? "Only Kimi, Gemini, MiniMax, and MiMo-V2.5 can analyze video. Switch to one of those via the model selector.",
         ...requestId ? { requestId } : {}
       };
     case "ezcoder":
       return {
-        headline: "ezcoder hit an unexpected error.",
+        headline: "EZ Coder hit an unexpected error.",
         source,
         message,
-        guidance: hint ?? "This looks like a ezcoder bug \u2014 please report it to the developer (see /help).",
+        guidance: hint ?? "This looks like a EZ Coder bug \u2014 please report it to the developer (see /help).",
         ...requestId ? { requestId } : {}
       };
   }
@@ -277,36 +277,39 @@ function providerGuidance(provider, message, statusCode) {
   const status = provider ? PROVIDER_STATUS_URL[provider] : void 0;
   const lower = message.toLowerCase();
   if (statusCode === 401 || lower.includes("unauthorized") || lower.includes("invalid api key")) {
-    return `Authentication failed with ${name}. Run \`ezcoder login\` to refresh your credentials.`;
+    return `Authentication failed with ${name}. Re-authenticate to refresh your credentials.`;
   }
   if (lower.includes("overloaded") || lower.includes("engine_overloaded")) {
-    return `${name}'s servers are overloaded right now. Retry in a moment \u2014 not a ezcoder issue.`;
+    return `${name}'s servers are overloaded right now. Retry in a moment \u2014 not a EZ Coder issue.`;
   }
   if (lower.includes("insufficient balance") || lower.includes("quota exceeded") || lower.includes("recharge") || lower.includes("no resource package")) {
-    return `Your ${name} account has a billing or quota issue \u2014 check your balance. Not a ezcoder issue.`;
+    return `Your ${name} account has a billing or quota issue \u2014 check your balance. Not a EZ Coder issue.`;
   }
   if (statusCode === 429 || lower.includes("rate limit") || lower.includes("too many requests")) {
-    return `${name} rate limit hit. Wait a moment then retry \u2014 not a ezcoder issue.`;
+    return `${name} rate limit hit. Wait a moment then retry \u2014 not a EZ Coder issue.`;
   }
   if (statusCode === 502 || lower.includes("bad gateway")) {
-    return `${name} returned a bad gateway. Retry \u2014 this is on their side, not ezcoder.`;
+    return `${name} returned a bad gateway. Retry \u2014 this is on their side, not EZ Coder.`;
   }
   if (statusCode === 503 || lower.includes("service unavailable")) {
-    return `${name} is temporarily unavailable. Retry shortly \u2014 not a ezcoder issue.`;
+    return `${name} is temporarily unavailable. Retry shortly \u2014 not a EZ Coder issue.`;
   }
   if (statusCode === 500 || lower.includes("server_error") || lower.includes("500") && lower.includes("internal server error")) {
-    return status ? `This is an error from ${name}, not ezcoder. Retry \u2014 if it keeps happening, check ${status}.` : `This is an error from ${name}, not ezcoder. Retry \u2014 if it keeps happening, try a different model with /model.`;
+    return status ? `This is an error from ${name}, not EZ Coder. Retry \u2014 if it keeps happening, check ${status}.` : `This is an error from ${name}, not EZ Coder. Retry \u2014 if it keeps happening, try a different model via the model selector.`;
   }
   if (lower.includes("timeout") || lower.includes("timed out")) {
-    return `Request to ${name} timed out. Their servers may be slow \u2014 retry. Not a ezcoder issue.`;
+    return `Request to ${name} timed out. Their servers may be slow \u2014 retry. Not a EZ Coder issue.`;
   }
   if (lower.includes("does not recognize the requested model") || lower.includes("model") && (lower.includes("not exist") || lower.includes("not found") || lower.includes("no access"))) {
-    return `${name} doesn't recognise this model on your account. Use /model to switch, or check your subscription tier.`;
+    return `${name} doesn't recognise this model on your account. Switch to a different model via the model selector, or check your subscription tier.`;
   }
   if (lower.includes("context_length_exceeded") || lower.includes("prompt is too long")) {
-    return `Context window for this ${name} model is full. Run /compact to shrink history, or start a new session.`;
+    return `Context window for this ${name} model is full. Compact the conversation to shrink history, or start a new session.`;
   }
-  return status ? `This is an error from ${name}, not ezcoder. Retry \u2014 if it persists, check ${status}.` : `This is an error from ${name}, not ezcoder. Retry \u2014 if it persists, try a different model with /model.`;
+  if (statusCode === 413 || lower.includes("request_too_large") || lower.includes("request exceeds the maximum size")) {
+    return `The request to ${name} is too large. Compact the conversation to shrink history, or start a new session.`;
+  }
+  return status ? `This is an error from ${name}, not EZ Coder. Retry \u2014 if it persists, check ${status}.` : `This is an error from ${name}, not EZ Coder. Retry \u2014 if it persists, try a different model via the model selector.`;
 }
 
 // src/providers/anthropic.ts
@@ -1091,6 +1094,12 @@ function parseToolArguments(argsJson) {
 // src/providers/anthropic.ts
 var NON_STREAMING_TIMEOUT_MS = 60 * 60 * 1e3;
 var anthropicClientCache = /* @__PURE__ */ new Map();
+function fineGrainedToolStreamingEnabled() {
+  const raw = process.env.GG_FINE_GRAINED_TOOL_STREAMING ?? process.env.CLAUDE_CODE_ENABLE_FINE_GRAINED_TOOL_STREAMING;
+  if (!raw) return false;
+  const v = raw.trim().toLowerCase();
+  return v === "1" || v === "true" || v === "yes" || v === "on";
+}
 function createClient(options) {
   const isOAuth = options.apiKey?.startsWith("sk-ant-oat");
   const userAgent = isOAuth ? options.userAgent ?? "claude-cli/2.1.75 (external, cli)" : "";
@@ -1148,7 +1157,9 @@ async function prewarmAnthropicCache(options) {
     ] : system;
     const tools = options.tools?.length ? toAnthropicTools(options.tools, {
       cacheControl,
-      enableFineGrainedToolStreaming: true
+      // Keep the serialized tool bytes identical to runStream so the
+      // prewarmed prompt cache actually hits — both are gated by the flag.
+      enableFineGrainedToolStreaming: fineGrainedToolStreamingEnabled()
     }) : void 0;
     await client.messages.create(
       {
@@ -1228,7 +1239,7 @@ async function* runStream(options) {
         options.tools.filter((t) => !reservedServerNames.has(t.name)),
         {
           ...supportsFirstPartyToolExtras && cacheControl ? { cacheControl } : {},
-          ...supportsFirstPartyToolExtras ? { enableFineGrainedToolStreaming: true } : {}
+          ...supportsFirstPartyToolExtras && fineGrainedToolStreamingEnabled() ? { enableFineGrainedToolStreaming: true } : {}
         }
       ) : [];
       return {
@@ -1254,7 +1265,10 @@ async function* runStream(options) {
     ...isOAuth ? ["claude-code-20250219", "oauth-2025-04-20"] : [],
     ...options.compaction ? ["compact-2026-01-12"] : [],
     ...options.clearToolUses ? ["context-management-2025-06-27"] : [],
-    "fine-grained-tool-streaming-2025-05-14",
+    // Eager tool-input streaming beta — opt-in only (see
+    // fineGrainedToolStreamingEnabled). Off by default: the un-buffered stream
+    // truncates large tool payloads into malformed JSON → phantom empty calls.
+    ...fineGrainedToolStreamingEnabled() ? ["fine-grained-tool-streaming-2025-05-14"] : [],
     ...!hasAdaptiveThinking ? ["interleaved-thinking-2025-05-14"] : [],
     // The 1-h cache TTL (cacheRetention "long") is gated behind this beta. Without
     // it Anthropic silently ignores ttl:"1h" and falls back to the 5-min default,
@@ -1387,7 +1401,14 @@ async function* runStream(options) {
               try {
                 const parsed = JSON.parse(accum.argsJson);
                 args = isJsonObject(parsed) ? parsed : {};
-              } catch {
+              } catch (parseErr) {
+                const rawPartial = accum.argsJson;
+                const snippet = rawPartial.length > 200 ? `${rawPartial.slice(0, 200)}\u2026` : rawPartial;
+                throw new ProviderError(
+                  "anthropic",
+                  `Tool "${accum.toolName}" input JSON was truncated in the stream (${rawPartial.length} bytes): ${snippet}; ${parseErr.message}`,
+                  { cause: parseErr }
+                );
               }
             }
             const tc = {
@@ -1608,6 +1629,7 @@ function readUnifiedRateLimit(headers) {
   return { rejected: status === "rejected", ...resetsAt ? { resetsAt } : {} };
 }
 function toError(err) {
+  if (err instanceof ProviderError) return err;
   if (err instanceof import_sdk.default.APIError) {
     const errorBody = err.error;
     const nestedError = errorBody?.error;
@@ -2238,10 +2260,10 @@ async function* runStream3(options) {
       if (options.model === "gpt-5.5-pro") {
         hint = "Use gpt-5.5 instead. OpenAI's Codex model catalog does not list gpt-5.5-pro.";
       } else {
-        hint = "This model is not available through Codex for the authenticated account. Run /model and choose a model listed for OpenAI Codex, or check your Codex model picker/usage limits.";
+        hint = "This model is not available through Codex for the authenticated account. Switch to a model listed for OpenAI Codex via the model selector, or check your Codex usage limits.";
       }
     } else if (response.status === 404 && text.includes("does not exist")) {
-      hint = "This model is not in the current OpenAI Codex catalog for this account. Try gpt-5.5, gpt-5.4, gpt-5.4-mini, or gpt-5.3-codex.";
+      hint = "This model is not in the current OpenAI Codex catalog for this account. Switch to gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, or gpt-5.5 via the model selector.";
     }
     throw new ProviderError("openai", message, {
       statusCode: response.status,
@@ -2665,10 +2687,11 @@ var CODE_ASSIST_SUPPORTED_MODELS = /* @__PURE__ */ new Set([
   "gemini-3.1-pro-preview",
   "gemini-3.1-pro-preview-customtools",
   "gemini-3-flash-preview",
-  "gemini-3.1-flash-lite-preview",
+  "gemini-3.5-flash",
+  "gemini-3-flash",
+  "gemini-3.1-flash-lite",
   "gemini-2.5-pro",
   "gemini-2.5-flash",
-  "gemini-2.5-flash-lite",
   "gemma-4-31b-it",
   "gemma-4-26b-a4b-it"
 ]);
@@ -2685,6 +2708,18 @@ function getCodeAssistEndpoint(method) {
 function formatUnsupportedModelMessage(model) {
   return `Gemini OAuth is configured to use the Gemini Code Assist subscription endpoint only. That endpoint does not currently expose model "${model}".`;
 }
+var ACCOUNT_GATED_MODELS = /* @__PURE__ */ new Set([
+  "gemini-3-flash",
+  "gemini-3.5-flash",
+  "gemini-3.1-pro-preview",
+  "gemini-3.1-pro-preview-customtools"
+]);
+function accountGatedMessage(model) {
+  return `Your Google account isn't entitled to "${model}" over Gemini Code Assist OAuth, so the API reports it as not found. This is an account-access limit, not a ezcoder bug.`;
+}
+function accountGatedHint() {
+  return `Newer Gemini models (3.5 Flash, 3.1 Pro Preview) are available only to Code Assist Standard/Enterprise accounts with preview/GA access enabled by a cloud admin \u2014 free/personal accounts usually can't call them. Switch to Gemini 3.1 Flash Lite (it works on this account) with /model, or sign in with a Code Assist Standard/Enterprise account that has preview access.`;
+}
 function formatErrorMessage(status, body, model) {
   if (status === 404 && !CODE_ASSIST_SUPPORTED_MODELS.has(model)) {
     return `Gemini API error (404): ${body}
@@ -2995,7 +3030,8 @@ async function fetchCodeAssist(plan, options) {
     if (!response.ok) {
       const text = await response.text().catch(() => "");
       const quota = parseGeminiQuota(response.status, text);
-      let message = formatErrorMessage(response.status, text, options.model);
+      const accountGated = response.status === 404 && ACCOUNT_GATED_MODELS.has(options.model);
+      let message = accountGated ? accountGatedMessage(options.model) : formatErrorMessage(response.status, text, options.model);
       let resetsAt;
       if (quota?.exhausted) {
         message = `Gemini quota exhausted \u2014 usage limit reached. ${message}`;
@@ -3004,7 +3040,8 @@ async function fetchCodeAssist(plan, options) {
       }
       throw new ProviderError("gemini", message, {
         statusCode: response.status,
-        ...resetsAt !== void 0 ? { resetsAt } : {}
+        ...resetsAt !== void 0 ? { resetsAt } : {},
+        ...accountGated ? { hint: accountGatedHint() } : {}
       });
     }
     return response;