fix: restore coding history and quota feedback

2026-09-01 11:46:04 +08:00
parent 38f85f6b5e
commit d523b72cf8
14 changed files with 480 additions and 49 deletions
--- a/.project-docs/30-worklog/tasks/20260831-fix-history-quota-errors-6c2a91e4.md
+++ b/.project-docs/30-worklog/tasks/20260831-fix-history-quota-errors-6c2a91e4.md
@@ -0,0 +1,104 @@
+# Task: Fix Coding history loading and quota errors
+
+## Identity
+
+- Task ID: 20260831-fix-history-quota-errors-6c2a91e4
+- Mode: Feature
+- Branch: codex/20260831-fix-history-quota-errors-6c2a91e4-fix-history-quota-errors-6c2a91e4
+- Worktree: /Users/inmanx/Documents/makelore-fix-history-quota-errors-6c2a91e4
+- Base commit: 38f85f6b5e4dc4e2c5e5b9f8f4506554cfd578f5
+- Owner: codex
+- Status: Ready for integration
+
+## Scope
+
+- Restore access to the full persisted active-branch history of Makelore Code
+  conversations after Pi compaction, while keeping history display bounded in
+  the Renderer and preserving the user's scroll anchor when older content is
+  prepended.
+- Recognize the observed upstream `403 token_point_balance_exhausted` contract
+  as a non-retryable Token Point quota failure and show actionable, safe
+  Chinese feedback instead of the generic `处理失败` summary.
+- Add focused Main/projector/Renderer regression coverage and run the required
+  typecheck, lint, and production build verification.
+
+## Intent And Constraints
+
+- Preserve Pi `0.84.2` as the sole runtime and keep Pi session/wire shapes
+  inside Electron Main. Renderer continues to consume product-neutral
+  Snapshot/Patch data through the Host API.
+- Full visible history must not change the compacted model context or expose
+  the private compaction summary. Accepted/uncertain mutations remain
+  non-replayed.
+- Quota classification must be narrow enough not to reinterpret unrelated
+  `403`, `429`, or Provider failures, and raw upstream text must not leak into
+  Renderer state.
+- The ready-for-integration progress-message task overlaps the timeline source
+  and test files but has no semantic conflict; keep changes localized to the
+  history-window and scrolling behavior.
+
+## Outcome
+
+- Pi session hydration now projects the complete active branch in chronological
+  order instead of dropping entries before the latest compaction boundary.
+  Compaction summaries remain private, and multiple durable compaction markers
+  reconcile against live IDs from newest to newest.
+- The Renderer still mounts only the latest 120 nodes initially, automatically
+  prepends earlier nodes in 100-node steps near the top, and offsets
+  `scrollTop` by the added height so the current reading position stays fixed.
+- The exact `403 token_point_balance_exhausted` / `词元点数余额不足` response is
+  classified as quota exhaustion, forwarded as non-retryable HTTP 402, and
+  projected as `CODING_PROVIDER_QUOTA_EXHAUSTED` with the safe message
+  `词元点数余额不足，请充值后重试。`.
+- Live failures and persisted Pi assistant failures share the same redacting
+  classifier. Reopened conversations restore only the safe Provider notice;
+  raw upstream request IDs are not copied into the public Snapshot.
+- The collapsed process summary displays the actionable public failure message
+  for both live runs and reopened persisted failures instead of generic
+  `处理失败`.
+
+## Verification
+
+- `pnpm exec vitest run tests/unit/pi-session-projector.test.ts tests/unit/coding-conversation-timeline.test.tsx tests/unit/ai-proxy-routes.test.ts tests/unit/pi-event-projector.test.ts`
+  - Passed: 4 files, 45 tests.
+- `pnpm run typecheck`
+  - Passed.
+- `pnpm run lint:check`
+  - Passed with 0 errors and 5 existing warnings in `src/pages/Home/index.tsx`
+    and `src/pages/Makelore/index.tsx`.
+- `pnpm test`
+  - Passed: 211 regular files / 1,730 tests plus the isolated pressure file /
+    1 test; 3 tests skipped.
+- `pnpm run build:vite`
+  - Passed for Renderer, Electron Main, Preload, and release utility worker;
+    existing Vite dynamic-import and chunk-size warnings remain.
+- `node ./node_modules/@playwright/test/cli.js test tests/e2e/pi-coding-first-chat.spec.ts`
+  - Passed: 2 Electron tests, including persisted quota feedback, automatic
+    older-history loading, and scroll-anchor preservation.
+- `git diff --check`
+  - Passed.
+
+## Follow-ups
+
+- Integration should account for the ready progress-message task because both
+  tasks intentionally touch `CodingConversationTimeline.tsx` and its unit/E2E
+  coverage; the edited concerns are separate.
+
+## Promotion Candidates
+
+- Target canonical documents: `README.md` and the integrated current-state
+  summary.
+- Proposal: state that Pi compaction changes model context but does not remove
+  visible active-branch history, and that recognized Provider quota failures
+  remain actionable after reopening a conversation.
+- Evidence: focused projector/Renderer/proxy tests, the 1,731-test full unit
+  run including pressure coverage, the production build, and the two focused
+  Electron E2E scenarios all pass.
+- Future impact: troubleshooting and future history work should preserve the
+  distinction between Pi context retention and product-visible session
+  history, and should reuse the redacted Provider-failure projection.
+- Semantic conflicts: none known; the peer progress-message task overlaps the
+  timeline files but not these semantics.
+- Human confirmation required: no product-direction choice is known to be
+  required, but the integration owner must decide whether this detail belongs
+  in canonical user-facing documentation.
--- a/electron/api/routes/ai-proxy.ts
+++ b/electron/api/routes/ai-proxy.ts
@@ -86,7 +86,8 @@ function getForwardedOneApiStatus(status: number, bodyText: string): number {
  if (status === 502 && isAIGatewayUserContextMissing(bodyText)) {
    return 401;
  }
-  if (status === 429 && getAIGatewayErrorKind(bodyText) === 'quota_exhausted') {
+  if ((status === 403 || status === 429)
+    && getAIGatewayErrorKind(bodyText) === 'quota_exhausted') {
    return 402;
  }
  if (status === 429 && isAIGatewayUpstreamSaturated(bodyText)) {
--- a/electron/coding-runtime/pi/event-projector.ts
+++ b/electron/coding-runtime/pi/event-projector.ts
@@ -6,8 +6,8 @@ import type {
  ConversationToolNode,
  PublicUsage,
 } from '../contracts';
-import { isAIGatewayUserContextMissing } from '../../../shared/ai-gateway-error-details';
 import type { PiRpcEvent } from './rpc-client';
+import { projectPiProviderFailure } from './provider-failure';
 import { subagentDetailsOfResult } from '../subagent-protocol';
 import {
  isProductToolName,
@@ -45,17 +45,6 @@ function retryFailure(message: string) {
  };
 }

-function providerFailure(message: unknown) {
-  if (typeof message === 'string' && isAIGatewayUserContextMissing(message)) {
-    return {
-      code: 'CODING_PROVIDER_AUTH_REQUIRED' as const,
-      message: '模型服务身份上下文无效，请重试；若仍失败请重新登录。',
-      recoverable: true,
-    };
-  }
-  return retryFailure('模型服务请求失败，请稍后重试。');
-}
-
 function asRecord(value: unknown): Record<string, unknown> | null {
  return value !== null && typeof value === 'object' && !Array.isArray(value)
    ? value as Record<string, unknown>
@@ -294,7 +283,7 @@ export class PiEventProjector {
          ...run,
          status: 'error',
          terminalReason: 'failed',
-          error: providerFailure(event.finalError),
+          error: projectPiProviderFailure(event.finalError),
        },
      }];
    }
@@ -485,7 +474,7 @@ export class PiEventProjector {
            ...snapshot.run,
            status: 'error',
            terminalReason: 'failed',
-            error: providerFailure(assistant.errorMessage),
+            error: projectPiProviderFailure(assistant.errorMessage),
          },
        }];
      }
--- a/electron/coding-runtime/pi/provider-failure.ts
+++ b/electron/coding-runtime/pi/provider-failure.ts
@@ -0,0 +1,25 @@
+import type { CodingRuntimePublicError } from '../contracts';
+import { isAIGatewayUserContextMissing } from '../../../shared/ai-gateway-error-details';
+import { getAIGatewayErrorKind } from '../../../shared/ai-gateway-error-kind';
+
+export function projectPiProviderFailure(message: unknown): CodingRuntimePublicError {
+  if (typeof message === 'string' && isAIGatewayUserContextMissing(message)) {
+    return {
+      code: 'CODING_PROVIDER_AUTH_REQUIRED',
+      message: '模型服务身份上下文无效，请重试；若仍失败请重新登录。',
+      recoverable: true,
+    };
+  }
+  if (typeof message === 'string' && getAIGatewayErrorKind(message) === 'quota_exhausted') {
+    return {
+      code: 'CODING_PROVIDER_QUOTA_EXHAUSTED',
+      message: '词元点数余额不足，请充值后重试。',
+      recoverable: false,
+    };
+  }
+  return {
+    code: 'CODING_RUNTIME_START_FAILED',
+    message: '模型服务请求失败，请稍后重试。',
+    recoverable: true,
+  };
+}
--- a/electron/coding-runtime/pi/session-projector.ts
+++ b/electron/coding-runtime/pi/session-projector.ts
@@ -14,6 +14,7 @@ import type {
 } from './event-projector';
 import { isProductToolName, productToolDetailsOfResult } from '../product-tool-protocol';
 import { subagentDetailsOfResult } from '../subagent-protocol';
+import { projectPiProviderFailure } from './provider-failure';

 export interface PiSessionSnapshotInput {
  snapshot: ConversationSnapshot;
@@ -106,23 +107,6 @@ function activePath(entriesValue: unknown, leafId: unknown): SessionEntryRecord[
  return path.reverse();
 }

-function retainedTail(path: SessionEntryRecord[]): SessionEntryRecord[] {
-  const compactionIndex = path.findLastIndex((entry) => entry.type === 'compaction');
-  if (compactionIndex < 0) return path;
-  const compaction = path[compactionIndex];
-  const firstKeptEntryId = compaction.firstKeptEntryId;
-  if (typeof firstKeptEntryId !== 'string') unreadable();
-  const firstKeptIndex = path.findIndex(
-    (entry, index) => index < compactionIndex && entry.id === firstKeptEntryId,
-  );
-  if (firstKeptIndex < 0) unreadable();
-  return [
-    compaction,
-    ...path.slice(firstKeptIndex, compactionIndex),
-    ...path.slice(compactionIndex + 1),
-  ];
-}
-
 async function contentBlocks(
  messageId: string,
  content: unknown,
@@ -265,6 +249,19 @@ async function projectEntries(
      ...(stopReason ? { stopReason } : {}),
    };
    nodes.push(node);
+    if (message.stopReason === 'error') {
+      const failure = projectPiProviderFailure(message.errorMessage);
+      if (failure.code === 'CODING_PROVIDER_AUTH_REQUIRED'
+        || failure.code === 'CODING_PROVIDER_QUOTA_EXHAUSTED') {
+        nodes.push({
+          kind: 'notice',
+          id: `${messageId}:provider-error`,
+          code: failure.code,
+          level: 'error',
+          message: failure.message,
+        });
+      }
+    }
    if (!Array.isArray(message.content)) continue;
    for (const value of message.content) {
      const block = asRecord(value);
@@ -310,6 +307,16 @@ function reconcileLiveIds(
  liveNodes: ConversationNode[],
 ): ConversationNode[] {
  const used = new Set<string>();
+  const durableCompactions = durableNodes.filter((node) => node.kind === 'compaction');
+  const liveCompactions = liveNodes.filter((node) => node.kind === 'compaction');
+  const compactionMatches = new Map<string, Extract<ConversationNode, { kind: 'compaction' }>>();
+  for (
+    let durableIndex = durableCompactions.length - 1, liveIndex = liveCompactions.length - 1;
+    durableIndex >= 0 && liveIndex >= 0;
+    durableIndex -= 1, liveIndex -= 1
+  ) {
+    compactionMatches.set(durableCompactions[durableIndex].id, liveCompactions[liveIndex]);
+  }
  return durableNodes.map((node) => {
    if (node.kind === 'message') {
      const live = liveNodes.find((candidate) => candidate.kind === 'message'
@@ -338,10 +345,8 @@ function reconcileLiveIds(
      };
    }
    if (node.kind === 'compaction') {
-      const live = liveNodes.findLast(
-        (candidate) => candidate.kind === 'compaction' && !used.has(candidate.id),
-      );
-      if (!live || live.kind !== 'compaction') return node;
+      const live = compactionMatches.get(node.id);
+      if (!live) return node;
      used.add(live.id);
      return { ...node, id: live.id, runId: live.runId };
    }
@@ -375,7 +380,7 @@ export async function projectPiSessionSnapshot(
  const response = asRecord(input.entries);
  const state = asRecord(input.state);
  if (!response || !state) unreadable();
-  const path = retainedTail(activePath(response.entries, response.leafId));
+  const path = activePath(response.entries, response.leafId);
  const durableNodes = await projectEntries(path, input);
  const nodes = reconcileLiveIds(durableNodes, input.snapshot.nodes);
  return {
--- a/shared/ai-gateway-error-kind.ts
+++ b/shared/ai-gateway-error-kind.ts
@@ -3,8 +3,10 @@ export type AIGatewayErrorKind = 'quota_exhausted' | 'authentication_invalid';
 export function isTokenBalanceExhausted(message: string | null | undefined): boolean {
  const normalized = message?.trim().toLowerCase() ?? '';
  return normalized.includes('token_balance_exhausted')
+    || normalized.includes('token_point_balance_exhausted')
    || normalized.includes('token balance exhausted')
-    || normalized.includes('balance exhausted');
+    || normalized.includes('balance exhausted')
+    || normalized.includes('词元点数余额不足');
 }

 export function getAIGatewayErrorKind(message: string | null | undefined): AIGatewayErrorKind | null {
--- a/shared/coding-conversation-contracts.ts
+++ b/shared/coding-conversation-contracts.ts
@@ -28,6 +28,7 @@ export type CodingRuntimeErrorCode =
  | 'CODING_RUNTIME_READY_TIMEOUT'
  | 'CODING_RUNTIME_PROTOCOL_ERROR'
  | 'CODING_PROVIDER_AUTH_REQUIRED'
+  | 'CODING_PROVIDER_QUOTA_EXHAUSTED'
  | 'CODING_MODEL_UNAVAILABLE'
  | 'CODING_SESSION_UNREADABLE'
  | 'CODING_STORAGE_WRITE_FAILED'
--- a/shared/coding-conversation-reducer.ts
+++ b/shared/coding-conversation-reducer.ts
@@ -60,6 +60,7 @@ const ERROR_CODES = new Set([
  'CODING_RUNTIME_READY_TIMEOUT',
  'CODING_RUNTIME_PROTOCOL_ERROR',
  'CODING_PROVIDER_AUTH_REQUIRED',
+  'CODING_PROVIDER_QUOTA_EXHAUSTED',
  'CODING_MODEL_UNAVAILABLE',
  'CODING_SESSION_UNREADABLE',
  'CODING_STORAGE_WRITE_FAILED',
--- a/src/pages/Chat/CodingConversationTimeline.tsx
+++ b/src/pages/Chat/CodingConversationTimeline.tsx
@@ -1265,6 +1265,19 @@ function processItemIsFailed(item: ProcessItem): boolean {
  return item.kind === 'notice' && item.node.level === 'error';
 }

+function persistedProviderFailureMessage(items: ProcessItem[]): string | undefined {
+  for (let index = items.length - 1; index >= 0; index -= 1) {
+    const item = items[index];
+    if (item?.kind === 'notice'
+      && item.node.level === 'error'
+      && (item.node.code === 'CODING_PROVIDER_AUTH_REQUIRED'
+        || item.node.code === 'CODING_PROVIDER_QUOTA_EXHAUSTED')) {
+      return item.node.message;
+    }
+  }
+  return undefined;
+}
+
 function processGroupIsFailed(
  items: ProcessItem[],
  run: ConversationRunState,
@@ -1326,10 +1339,13 @@ const ProcessGroup = memo(function ProcessGroup({
      ? runningDurationLabel(run, now)
      : settledDurationLabel(run)
    : null;
+  const failureMessage = failed
+    ? (latest ? run.error?.message : undefined) ?? persistedProviderFailureMessage(items)
+    : undefined;
  const summary = active
    ? `处理中${duration ? ` ${duration}` : ''}`
    : failed
-      ? `处理失败${duration ? ` · ${duration}` : ''}`
+      ? `${failureMessage ?? '处理失败'}${duration ? ` · ${duration}` : ''}`
      : `已处理${duration ? ` ${duration}` : ''}`;

  return (
@@ -1491,11 +1507,33 @@ export const CodingConversationTimeline = memo(function CodingConversationTimeli
  const nodes = useCodingConversationStore(selectNodes);
  const run = useCodingConversationStore(selectRun);
  const [visibleLimit, setVisibleLimit] = useState(INITIAL_WINDOW);
+  const scrollRef = useRef<HTMLDivElement | null>(null);
+  const stickToBottomRef = useRef(true);
+  const prependAnchorRef = useRef<{ scrollHeight: number; scrollTop: number } | null>(null);
  const windowStart = Math.max(0, nodes.length - visibleLimit);
  const visibleNodes = nodes.slice(windowStart);
  const turns = useMemo(() => timelineTurns(visibleNodes), [visibleNodes]);
-  const scrollRef = useRef<HTMLDivElement | null>(null);
-  const stickToBottomRef = useRef(true);
+
+  const loadEarlier = useCallback(() => {
+    if (windowStart === 0 || prependAnchorRef.current) return;
+    const element = scrollRef.current;
+    if (!element) return;
+    prependAnchorRef.current = {
+      scrollHeight: element.scrollHeight,
+      scrollTop: element.scrollTop,
+    };
+    stickToBottomRef.current = false;
+    setVisibleLimit((current) => Math.min(nodes.length, current + WINDOW_STEP));
+  }, [nodes.length, windowStart]);
+
+  useLayoutEffect(() => {
+    const anchor = prependAnchorRef.current;
+    if (!anchor) return;
+    prependAnchorRef.current = null;
+    const element = scrollRef.current;
+    if (!element) return;
+    element.scrollTop = anchor.scrollTop + (element.scrollHeight - anchor.scrollHeight);
+  }, [visibleLimit]);

  useLayoutEffect(() => {
    if (!stickToBottomRef.current) return;
@@ -1513,6 +1551,7 @@ export const CodingConversationTimeline = memo(function CodingConversationTimeli
      onScroll={(event) => {
        const element = event.currentTarget;
        stickToBottomRef.current = element.scrollHeight - element.scrollTop - element.clientHeight < 64;
+        if (element.scrollTop <= 48) loadEarlier();
      }}
    >
      <div className="mx-auto flex w-full max-w-[50rem] flex-col gap-8">
@@ -1521,7 +1560,7 @@ export const CodingConversationTimeline = memo(function CodingConversationTimeli
            type="button"
            variant="ghost"
            className="mx-auto min-h-9 rounded-lg px-4 text-xs transition-transform duration-150 ease-out active:scale-[0.96]"
-            onClick={() => setVisibleLimit((current) => current + WINDOW_STEP)}
+            onClick={loadEarlier}
          >
            加载更早内容
          </Button>
--- a/tests/e2e/pi-coding-first-chat.spec.ts
+++ b/tests/e2e/pi-coding-first-chat.spec.ts
@@ -143,6 +143,12 @@ async function installCodingFirstChatHost(
        : null,
      modelResolution: featureComplete ? 'resolved' : 'required',
    };
+    const historyConversation = {
+      ...conversation,
+      id: 'conversation-pi-history',
+      title: 'History and quota',
+      updatedAt: '2026-08-23T23:58:00.000Z',
+    };
    const snapshot = {
      schemaVersion: 1,
      conversation: {
@@ -353,6 +359,39 @@ async function installCodingFirstChatHost(
        error: { code: 'CODING_RUNTIME_START_FAILED', message: 'Worker stopped', recoverable: true },
      },
    };
+    const historySnapshot = {
+      ...snapshot,
+      conversation: {
+        ...snapshot.conversation,
+        id: historyConversation.id,
+        title: historyConversation.title,
+      },
+      nodes: [
+        ...Array.from({ length: 150 }, (_, index) => ({
+          kind: 'message',
+          id: `history-message-${index}`,
+          role: 'user',
+          status: 'complete',
+          blocks: [{
+            kind: 'text',
+            id: `history-message-${index}:content:0`,
+            text: `History message ${index}`,
+            status: 'complete',
+          }],
+        })),
+        {
+          kind: 'notice',
+          id: 'history-quota-notice',
+          code: 'CODING_PROVIDER_QUOTA_EXHAUSTED',
+          level: 'error',
+          message: '词元点数余额不足，请充值后重试。',
+        },
+      ],
+      run: { status: 'idle' },
+      queue: { items: [] },
+      pendingInteractions: [],
+      worker: { status: 'ready', generation: 1 },
+    };
    const respond = (json: unknown, status = 200) => ({
      ok: true,
      data: { status, ok: status >= 200 && status < 300, json },
@@ -453,7 +492,7 @@ async function installCodingFirstChatHost(
      if (path === `/api/coding/projects/conversations?projectId=${project.id}`) {
        return respond({
          conversations: featureComplete
-            ? [conversation, secondConversation]
+            ? [conversation, secondConversation, historyConversation]
            : state.conversationCreated
              ? [conversation]
              : [],
@@ -478,6 +517,9 @@ async function installCodingFirstChatHost(
      if (path === `/api/coding/conversations/${secondConversation.id}/snapshot`) {
        return respond({ snapshot: secondSnapshot });
      }
+      if (path === `/api/coding/conversations/${historyConversation.id}/snapshot`) {
+        return respond({ snapshot: historySnapshot });
+      }
      if (path === `/api/coding/conversations/${conversation.id}/prompt` && method === 'POST') {
        return respond({
          acceptance: {
@@ -890,6 +932,29 @@ test('PI feature UI isolates Conversations and exposes queue, interaction, model
    await expect(page.getByRole('textbox')).toHaveValue('');
    await page.getByRole('button', { name: '恢复' }).click();

+    await builderConversations.getByRole('button', { name: 'History and quota' }).click();
+    const historyProcess = page.getByTestId('coding-process-group');
+    await expect(historyProcess.locator('summary').first())
+      .toContainText('词元点数余额不足，请充值后重试。');
+    await expect(page.getByText('History message 0')).toHaveCount(0);
+    await expect(page.getByText('History message 31')).toHaveCount(1);
+    const historyTimeline = page.getByTestId('coding-conversation-timeline');
+    const beforeHeight = await historyTimeline.evaluate((element) => {
+      element.scrollTop = 0;
+      const height = element.scrollHeight;
+      element.dispatchEvent(new Event('scroll', { bubbles: true }));
+      return height;
+    });
+    await expect(page.getByText('History message 0')).toHaveCount(1);
+    const anchoredScroll = await historyTimeline.evaluate((element) => ({
+      scrollHeight: element.scrollHeight,
+      scrollTop: element.scrollTop,
+    }));
+    expect(anchoredScroll.scrollTop).toBeGreaterThan(0);
+    expect(Math.abs(
+      anchoredScroll.scrollTop - (anchoredScroll.scrollHeight - beforeHeight),
+    )).toBeLessThanOrEqual(2);
+
    await expect(page.getByText(/编程工具|分享|取消分享|回滚|恢复回滚|待办|全局运行时/)).toHaveCount(0);
    const state = await readState(electronApp);
    expect(state.captured.some((request) => request.path.endsWith('/model') && request.method === 'POST')).toBe(true);
--- a/tests/unit/ai-proxy-routes.test.ts
+++ b/tests/unit/ai-proxy-routes.test.ts
@@ -274,6 +274,40 @@ describe('ai proxy routes', () => {
    expect(response.body()).toBe(JSON.stringify({ error: 'user quota is not enough' }));
  });

+  it('maps the observed token-point balance 403 to a non-retryable quota response', async () => {
+    seedWorksSquareAIGatewayCredential({
+      accessToken: 'ws-ai-token',
+      expiresIn: 3600,
+      oneApiBaseUrl: 'https://one-api.example.com/v1',
+    });
+    const body = JSON.stringify({
+      error: {
+        message: '词元点数余额不足',
+        code: 'token_point_balance_exhausted',
+        type: 'one_api_error',
+      },
+    });
+    const fetchMock = vi.fn().mockResolvedValueOnce(
+      new Response(body, {
+        status: 403,
+        headers: { 'content-type': 'application/json' },
+      }),
+    );
+    vi.stubGlobal('fetch', fetchMock);
+    const response = createResponse();
+
+    await handleAiProxyRoutes(
+      createRequest('POST', { model: 'deepseek-v4-flash', messages: [] }),
+      response.res,
+      new URL('http://127.0.0.1:13210/api/ai-proxy/v1/chat/completions'),
+      {} as never,
+    );
+
+    expect(fetchMock).toHaveBeenCalledOnce();
+    expect(response.statusCode).toBe(402);
+    expect(response.body()).toBe(body);
+  });
+
  it('maps rolling-window quota exhaustion to a non-retryable response status', async () => {
    seedWorksSquareAIGatewayCredential({
      accessToken: 'ws-ai-token',
--- a/tests/unit/coding-conversation-timeline.test.tsx
+++ b/tests/unit/coding-conversation-timeline.test.tsx
@@ -631,7 +631,83 @@ describe('CodingConversationTimeline', () => {
      .toBeVisible();
  });

-  it('windows long timelines and loads earlier nodes only on demand', async () => {
+  it('shows an actionable non-retryable quota failure in the process summary', async () => {
+    const { codingConversationStore } = await import('@/stores/coding-conversations');
+    const { CodingConversationTimeline } = await import(
+      '@/pages/Chat/CodingConversationTimeline'
+    );
+    const base = createProductSnapshot('conversation-quota-failure', 1);
+    const snapshot = {
+      ...base,
+      run: {
+        status: 'idle' as const,
+        runId: 'run-quota-failure',
+        startedAt: 1_000,
+        settledAt: 6_000,
+        terminalReason: 'failed' as const,
+        error: {
+          code: 'CODING_PROVIDER_QUOTA_EXHAUSTED' as const,
+          message: '词元点数余额不足，请充值后重试。',
+          recoverable: false,
+        },
+      },
+      nodes: [{
+        kind: 'tool' as const,
+        id: 'tool-quota-failure',
+        toolCallId: 'call-quota-failure',
+        toolName: 'bash',
+        title: '执行命令',
+        inputText: '',
+        status: 'error' as const,
+        output: [],
+      }],
+    };
+    codingConversationStore.getState().applySnapshotEvent({
+      type: 'snapshot',
+      conversationId: 'conversation-quota-failure',
+      workerGeneration: 1,
+      seq: snapshot.cursor.seq,
+      snapshot,
+    });
+
+    render(<CodingConversationTimeline conversationId="conversation-quota-failure" />);
+
+    expect(within(screen.getByTestId('coding-process-group'))
+      .getByText('词元点数余额不足，请充值后重试。 · 5 秒')).toBeVisible();
+  });
+
+  it('shows an actionable persisted Provider failure after reopening a Conversation', async () => {
+    const { codingConversationStore } = await import('@/stores/coding-conversations');
+    const { CodingConversationTimeline } = await import(
+      '@/pages/Chat/CodingConversationTimeline'
+    );
+    const base = createProductSnapshot('conversation-persisted-quota-failure', 1);
+    const snapshot = {
+      ...base,
+      run: { status: 'idle' as const },
+      nodes: [{
+        kind: 'notice' as const,
+        id: 'persisted-quota-notice',
+        code: 'CODING_PROVIDER_QUOTA_EXHAUSTED',
+        level: 'error' as const,
+        message: '词元点数余额不足，请充值后重试。',
+      }],
+    };
+    codingConversationStore.getState().applySnapshotEvent({
+      type: 'snapshot',
+      conversationId: 'conversation-persisted-quota-failure',
+      workerGeneration: 1,
+      seq: snapshot.cursor.seq,
+      snapshot,
+    });
+
+    render(<CodingConversationTimeline conversationId="conversation-persisted-quota-failure" />);
+
+    expect(screen.getByTestId('coding-process-group').querySelector('summary'))
+      .toHaveTextContent('词元点数余额不足，请充值后重试。');
+  });
+
+  it('windows long timelines, loads near the top, and preserves the scroll anchor', async () => {
    const { codingConversationStore } = await import('@/stores/coding-conversations');
    const { CodingConversationTimeline } = await import(
      '@/pages/Chat/CodingConversationTimeline'
@@ -659,8 +735,17 @@ describe('CodingConversationTimeline', () => {

    expect(screen.queryByText('Notice 0')).not.toBeInTheDocument();
    expect(screen.getByText('Notice 30')).toBeInTheDocument();
-    fireEvent.click(screen.getByRole('button', { name: '加载更早内容' }));
-    expect(screen.getByText('Notice 0')).toBeInTheDocument();
+    const timeline = screen.getByTestId('coding-conversation-timeline');
+    Object.defineProperty(timeline, 'scrollHeight', {
+      configurable: true,
+      get: () => screen.queryByText('Notice 0') ? 1_600 : 1_000,
+    });
+    timeline.scrollTop = 0;
+
+    fireEvent.scroll(timeline);
+
+    await waitFor(() => expect(screen.getByText('Notice 0')).toBeInTheDocument());
+    expect(timeline.scrollTop).toBe(600);
  });

  it('does not commit the selected timeline when a hidden Conversation streams', async () => {
--- a/tests/unit/pi-event-projector.test.ts
+++ b/tests/unit/pi-event-projector.test.ts
@@ -472,6 +472,40 @@ describe('Pi event projector', () => {
    expect(JSON.stringify(snapshot)).not.toContain('secret request detail');
  });

+  it('projects token-point balance exhaustion as a non-retryable Provider quota failure', async () => {
+    const projector = new PiEventProjector({ createId: () => 'provider-quota-error-a' });
+    let snapshot = emptySnapshot();
+    const providerFailure = {
+      role: 'assistant',
+      content: [],
+      stopReason: 'error',
+      errorMessage: '403: {"error":{"message":"词元点数余额不足","code":"token_point_balance_exhausted","request_id":"secret-request"}}',
+      timestamp: 10,
+    };
+
+    snapshot = apply(snapshot, await projector.project(snapshot, {
+      type: 'message_end',
+      message: providerFailure,
+    }));
+    snapshot = apply(snapshot, await projector.project(snapshot, {
+      type: 'agent_end',
+      messages: [providerFailure],
+      willRetry: false,
+    }));
+    snapshot = apply(snapshot, await projector.project(snapshot, { type: 'agent_settled' }));
+
+    expect(snapshot.run).toMatchObject({
+      status: 'idle',
+      terminalReason: 'failed',
+      error: {
+        code: 'CODING_PROVIDER_QUOTA_EXHAUSTED',
+        message: '词元点数余额不足，请充值后重试。',
+        recoverable: false,
+      },
+    });
+    expect(JSON.stringify(snapshot)).not.toContain('secret-request');
+  });
+
  it('keeps an exhausted Works user-context retry redacted and Provider-owned', async () => {
    const projector = new PiEventProjector({ createId: () => 'provider-retry-error-a' });
    let snapshot = emptySnapshot();
--- a/tests/unit/pi-session-projector.test.ts
+++ b/tests/unit/pi-session-projector.test.ts
@@ -242,7 +242,7 @@ describe('Pi session projector', () => {
    expect(snapshot.cursor).toEqual({ workerGeneration: 1, seq: 7 });
  });

-  it('applies retained-tail compaction and reconciles durable entries without replacing live IDs', async () => {
+  it('projects the full active history around compaction and reconciles durable entries without replacing live IDs', async () => {
    const live: ConversationSnapshot = {
      ...baseSnapshot(),
      nodes: [
@@ -371,10 +371,11 @@ describe('Pi session projector', () => {
    });

    expect(snapshot.nodes.map(({ kind, id }) => ({ kind, id }))).toEqual([
-      { kind: 'compaction', id: 'live-compaction-a' },
+      { kind: 'message', id: 'entry:entry-old-user' },
      { kind: 'message', id: 'live-user-a' },
      { kind: 'message', id: 'live-assistant-a' },
      { kind: 'tool', id: 'live-tool-a' },
+      { kind: 'compaction', id: 'live-compaction-a' },
      { kind: 'message', id: 'entry:entry-after-user' },
    ]);
    expect(snapshot.nodes).toContainEqual(expect.objectContaining({
@@ -401,10 +402,55 @@ describe('Pi session projector', () => {
      compaction: 'idle',
      recalculating: true,
    });
-    expect(JSON.stringify(snapshot)).not.toContain('Summarized old question');
+    expect(JSON.stringify(snapshot)).toContain('Summarized old question');
    expect(JSON.stringify(snapshot)).not.toContain('summary must stay hidden');
  });

+  it('restores persisted Provider quota failures as safe actionable notices', async () => {
+    const snapshot = await projectPiSessionSnapshot({
+      snapshot: baseSnapshot(),
+      workerGeneration: 2,
+      state: { sessionId: 'pi-session-a', isStreaming: false, isCompacting: false },
+      entries: {
+        leafId: 'entry-quota-error',
+        entries: [
+          {
+            type: 'message',
+            id: 'entry-quota-user',
+            parentId: null,
+            message: { role: 'user', content: 'Continue', timestamp: 1 },
+          },
+          {
+            type: 'message',
+            id: 'entry-quota-error',
+            parentId: 'entry-quota-user',
+            message: {
+              role: 'assistant',
+              content: [],
+              stopReason: 'error',
+              errorMessage: '403: {"message":"词元点数余额不足","code":"token_point_balance_exhausted","request_id":"secret-request"}',
+              timestamp: 2,
+            },
+          },
+        ],
+      },
+    });
+
+    expect(snapshot.nodes).toContainEqual(expect.objectContaining({
+      kind: 'message',
+      id: 'entry:entry-quota-error',
+      status: 'error',
+    }));
+    expect(snapshot.nodes).toContainEqual({
+      kind: 'notice',
+      id: 'entry:entry-quota-error:provider-error',
+      code: 'CODING_PROVIDER_QUOTA_EXHAUSTED',
+      level: 'error',
+      message: '词元点数余额不足，请充值后重试。',
+    });
+    expect(JSON.stringify(snapshot)).not.toContain('secret-request');
+  });
+
  it('projects persisted images through attachment storage without retaining base64', async () => {
    const rawImage = 'A'.repeat(1024 * 1024);
    const projected = await projectPiSessionSnapshot({