@hienlh/ppm 0.13.93 → 0.13.95

package/src/providers/claude-agent-sdk.ts

@@ -833,6 +833,15 @@ export class ClaudeAgentSdkProvider implements AIProvider {
         console.log(`[sdk] session=${sessionId} mcpServers: ${Object.keys(mcpServers).join(", ")}`);
       }
 
+      // 1M context (GA): the CLI enables a 1M window when the model name carries a
+      // [1m] suffix. The suffix is stripped before the API call. Requires an entitled
+      // account (Max/Team/Enterprise) and a supported model; otherwise the API errors.
+      const baseModel = opts?.model ?? providerConfig.model;
+      const resolvedModel =
+        baseModel && providerConfig.context_1m && !/\[1m\]$/i.test(baseModel)
+          ? `${baseModel}[1m]`
+          : baseModel;
+
       const queryOptions: Record<string, any> = {
         // On Windows, child_process.spawn("bun") fails with ENOENT — force node
         ...(process.platform === "win32" && { executable: "node" }),
@@ -849,13 +858,14 @@ export class ClaudeAgentSdkProvider implements AIProvider {
         ...(hasMcp && { mcpServers }),
         permissionMode,
         allowDangerouslySkipPermissions: isBypass,
-        ...((opts?.model ?? providerConfig.model) && { model: opts?.model ?? providerConfig.model }),
+        ...(resolvedModel && { model: resolvedModel }),
         ...(providerConfig.effort && { effort: providerConfig.effort }),
         maxTurns: providerConfig.max_turns ?? 1000,
         ...(providerConfig.max_budget_usd && { maxBudgetUsd: providerConfig.max_budget_usd }),
         ...(providerConfig.thinking_budget_tokens != null && {
           maxThinkingTokens: providerConfig.thinking_budget_tokens,
         }),
+        ...(providerConfig.context_1m && { betas: ["context-1m-2025-08-07"] }),
         includePartialMessages: true,
         stderr: stderrCallback,
       };
@@ -875,18 +885,28 @@ export class ClaudeAgentSdkProvider implements AIProvider {
         session_id: sessionId,
       };
 
-      // Build a retry message from the LATEST user content (not the stale firstMsg).
-      // Follow-ups via pushMessage update lastUserContent on the streaming session,
-      // so retries after token refresh correctly replay the current turn.
+      // Once the turn has produced output (top-level assistant text/tool_use), the user
+      // message + any tool_results are persisted to JSONL. Mid-turn retries must resume and
+      // continue — NOT re-push the original message, which restarts the turn and re-displays
+      // an already-answered AskUserQuestion. Persists across retryLoop iterations.
+      let turnProgressed = false;
+
+      // Build a retry message for token-refresh / transient-error retries.
+      // Pre-turn (nothing persisted yet): re-push the LATEST user content so the SDK has a
+      //   prompt (follow-ups via pushMessage keep lastUserContent current).
+      // Mid-turn (turnProgressed): the SDK's resume loads the pending tool_result and marks
+      //   the turn `interrupted_turn`; we push a continuation nudge to drive the generator so
+      //   it picks up where it stopped instead of duplicating the turn.
       // Also returns the raw content/images for re-populating the new streaming session.
       const buildRetryMsg = () => {
         const ss = this.streamingSessions.get(sessionId);
         const content = ss?.lastUserContent ?? message;
         const images = ss?.lastUserImages;
+        const retryContent = turnProgressed ? "Continue from where you left off." : content;
         return {
           msg: {
             type: 'user' as const,
-            message: buildMessageParam(content, images),
+            message: buildMessageParam(retryContent, turnProgressed ? undefined : images),
             parent_tool_use_id: null,
             session_id: sessionId,
           },
@@ -1165,12 +1185,20 @@ export class ClaudeAgentSdkProvider implements AIProvider {
           // Detect 401 pattern in text: "Failed to authenticate. API Error: 401 ..."
           if (!assistantError) {
             const textContent = this.extractAssistantText(msg);
-            if (textContent && /API Error:\s*401\b.*authentication_error/i.test(textContent)) {
+            if (textContent && /API Error:\s*401\b/i.test(textContent)) {
               assistantError = "authentication_failed";
               console.warn(`[sdk] session=${sessionId} detected 401 in assistant text content — treating as auth error`);
             } else if (textContent && /hit your (?:[\w-]+\s+)*limit/i.test(textContent)) {
               assistantError = "rate_limit";
               console.warn(`[sdk] session=${sessionId} detected quota limit in assistant text content — treating as rate_limit`);
+            } else if (textContent && /API Error:\s*5\d{2}\b/i.test(textContent)) {
+              // 5xx (e.g. 529 Overloaded) — match the explicit "API Error: 5xx" text only.
+              // Treat as server_error so it enters the retry branch and the raw error text is
+              // not streamed as duplicated assistant content. Deliberately NOT keyed off
+              // isApiErrorMessage alone — that flag is also set for non-retryable 4xx errors
+              // (e.g. billing/invalid_request), which must surface immediately, not retry.
+              assistantError = "server_error";
+              console.warn(`[sdk] session=${sessionId} detected API 5xx server error — treating as server_error`);
             }
           }
 
@@ -1228,7 +1256,9 @@ export class ClaudeAgentSdkProvider implements AIProvider {
               } else {
                 console.warn(`[sdk] session=${sessionId} rate limited — retrying in ${backoff / 1000}s (attempt ${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})`);
               }
-              yield { type: "error", message: `Rate limited. Auto-retrying in ${backoff / 1000}s... (${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})` };
+              // Replaceable status line (not an `error` event) so repeated retries don't
+              // accumulate duplicate error blocks — only the final error (if any) is shown.
+              yield { type: "status_update", phase: "retrying", message: `Đang thử lại (${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})...` };
               await new Promise((r) => setTimeout(r, backoff));
               // Close current streaming session and recreate with (potentially new) account env.
               const retry4 = buildRetryMsg();
@@ -1260,6 +1290,9 @@ export class ClaudeAgentSdkProvider implements AIProvider {
             // Skip emitting the raw 401 error as text content — already shown as error event
             continue;
           }
+          // Successful top-level assistant output → turn is now persisted to JSONL.
+          // Subsequent retries this turn must continue, not re-push the original message.
+          if (!parentId) turnProgressed = true;
           const content = (msg as any).message?.content;
           if (Array.isArray(content)) {
             for (const block of content) {
@@ -1325,7 +1358,9 @@ export class ClaudeAgentSdkProvider implements AIProvider {
                 } else {
                   console.warn(`[sdk] session=${sessionId} result 429 — retrying in ${backoff / 1000}s (attempt ${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})`);
                 }
-                yield { type: "error", message: `Rate limited. Auto-retrying in ${backoff / 1000}s... (${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})` };
+                // Replaceable status line (not an `error` event) — avoids accumulating
+                // duplicate error blocks across retries; only a final error is shown.
+                yield { type: "status_update", phase: "retrying", message: `Đang thử lại (${rateLimitRetryCount}/${MAX_RATE_LIMIT_RETRIES})...` };
                 await new Promise((r) => setTimeout(r, backoff));
                 const retry5 = buildRetryMsg();
                 closeCurrentStream();

package/src/services/jsonl-transcript-parser.ts

@@ -49,7 +49,19 @@ export function parseSessionMessage(
         (b) => b.type === "text" && typeof b.text === "string" &&
           /Failed to authenticate|API Error: 40[13]|hit your limit|rate.?limit/i.test(b.text as string),
       ));
-  if (isSdkErrorMessage) {
+  // SDK emits a placeholder "No response requested." assistant turn after some
+  // interrupted/empty turns (e.g. following a 5xx/Overloaded failure). It carries no
+  // value — drop so it doesn't render as a stray bubble when reloading the session.
+  const isNoOpAssistant =
+    role === "assistant" &&
+    Array.isArray(message?.content) &&
+    (message!.content as Array<Record<string, unknown>>).length > 0 &&
+    (message!.content as Array<Record<string, unknown>>).every(
+      (b) => b.type === "text" && typeof b.text === "string" &&
+        (b.text as string).trim() === "No response requested.",
+    );
+
+  if (isSdkErrorMessage || isNoOpAssistant) {
     return {
       id: msg.uuid,
       role,

package/src/types/chat.ts

@@ -134,7 +134,7 @@ export type ChatEvent =
   | { type: "done"; sessionId: string; resultSubtype?: ResultSubtype; numTurns?: number; contextWindowPct?: number; lastMessageUuid?: string }
   | { type: "account_info"; accountId: string; accountLabel: string }
   | { type: "account_retry"; reason: string; accountId?: string; accountLabel?: string }
-  | { type: "status_update"; phase: "routing" | "refreshing" | "switching"; message: string; accountLabel?: string }
+  | { type: "status_update"; phase: "routing" | "refreshing" | "switching" | "retrying"; message: string; accountLabel?: string }
   | { type: "system"; subtype: string }
   | { type: "team_detected"; teamName: string }
   | { type: "team_updated"; teamName: string; team: unknown }

package/src/types/config.ts

@@ -63,6 +63,9 @@ export interface AIProviderConfig {
   max_budget_usd?: number;
   thinking_budget_tokens?: number;
   agent_teams?: boolean;
+  // Enable 1M context window via beta header. Requires an entitled account
+  // (Max/Team/Enterprise) and an opus-4/sonnet-4 model; otherwise the API errors.
+  context_1m?: boolean;
 
   // CLI-specific (Cursor, Codex, Gemini)
   cli_command?: string;

package/src/web/components/settings/ai-settings-section.tsx

@@ -285,6 +285,20 @@ export function AISettingsSection({ compact }: { compact?: boolean } = {}) {
             {config?.agent_teams && (
               <TeamListSection compact={compact} />
             )}
+
+            <div className="flex items-center justify-between gap-2">
+              <div>
+                <Label htmlFor="ai-context-1m" className={compact ? labelSize : undefined}>1M Context Window</Label>
+                <p className={`${compact ? "text-[9px]" : "text-[11px]"} text-muted-foreground`}>
+                  Requires an entitled account (Max/Team/Enterprise) and an Opus 4 / Sonnet 4 model. Other accounts will error.
+                </p>
+              </div>
+              <Switch
+                id="ai-context-1m"
+                checked={config?.context_1m ?? false}
+                onCheckedChange={(v) => handleSave("context_1m", v)}
+              />
+            </div>
           </>
         )}
 

package/src/web/hooks/use-chat.ts

@@ -337,13 +337,20 @@ export function useChat(sessionId: string | null, providerId = "claude", project
       }
 
       case "error": {
-        streamingEventsRef.current.push(ev as ChatEvent);
-        const errEvents = [...streamingEventsRef.current];
+        // Safety net: dedupe identical consecutive errors (e.g. repeated 5xx) so a turn
+        // never accumulates duplicate error blocks — keep only one.
+        const evs = streamingEventsRef.current;
+        const lastErr = evs[evs.length - 1];
+        const isDupErr = lastErr?.type === "error" && (lastErr as { message?: string }).message === ev.message;
+        if (!isDupErr) evs.push(ev as ChatEvent);
+        const errEvents = [...evs];
         setMessages((prev) => {
           const last = prev[prev.length - 1];
           if (last?.role === "assistant") {
             return [...prev.slice(0, -1), { ...last, events: errEvents }];
           }
+          // System-message fallback: skip if the trailing system message is the same error.
+          if (last?.role === "system" && last.content === ev.message) return prev;
           return [...prev, {
             id: `error-${Date.now()}`,
             role: "system" as const,

package/src/web/lib/api-settings.ts

@@ -179,6 +179,7 @@ export interface AIProviderSettings {
   permission_mode?: string;
   system_prompt?: string;
   agent_teams?: boolean;
+  context_1m?: boolean;
 }
 
 export interface AISettings {

package/test-tool.mjs

@@ -1,5 +1,10 @@
 import { tmpdir } from "node:os";
 import { ClaudeAgentSdkProvider } from "./src/providers/claude-agent-sdk.ts";
+import { configService } from "./src/services/config.service.ts";
+
+// Load real config from SQLite (provider reads model/context_1m from here).
+// Without this, configService falls back to DEFAULT_CONFIG (sonnet-4-6, no 1M).
+configService.load();
 
 // Remove CLAUDECODE to avoid nested session error
 delete process.env.CLAUDECODE;

package/dist/web/assets/keybindings-store-vRxA7Pu1.js

@@ -1 +0,0 @@
-import"./vendor-markdown-0Mxgxy0L.js";import"./api-client-bbJLzRVE.js";import{L as e}from"./index-TZXL49LV.js";export{e as useKeybindingsStore};

package/dist/web/assets/notification-store-C5AT0vIG.js

@@ -1 +0,0 @@
-import"./vendor-markdown-0Mxgxy0L.js";import"./api-client-bbJLzRVE.js";import{W as e}from"./index-TZXL49LV.js";export{e as useNotificationStore};