fix: resolve desktop startup and provider regressions

Fixes #827, #857, #860, #885.

Reproduced #827 with a delayed pasted-image FileReader result that reappeared after send; fixed by invalidating stale paste callbacks after submit or session switch.

Reproduced #857 with TCP accepting while /health stayed unhealthy; fixed Electron sidecar readiness to require HTTP /health status ok before continuing.

Reproduced #860 through OpenAI-compatible sampling param pass-through and missing GLM-5.2 context metadata; fixed sampling params to be opt-in and added GLM-5.2 context windows for preset and core resolution.

Reproduced #885 with a truncated long sidebar title lacking a row tooltip; fixed by exposing the full title on the session row.

Tested: cd desktop && bun run test -- src/components/chat/ChatInput.test.tsx src/components/layout/Sidebar.test.tsx --run

Tested: cd desktop && bun test electron/services/sidecarManager.test.ts

Tested: bun test src/server/__tests__/proxy-transform.test.ts src/server/__tests__/provider-presets.test.ts src/services/compact/autoCompact.test.ts

Tested: bun run check:server

Tested: bun run check:desktop

Tested: bun run check:native

Confidence: high

Scope-risk: moderate

desktop/electron/services/sidecarManager.test.ts

@@ -1,5 +1,6 @@
 import { describe, expect, it, vi } from 'vitest'
 import net from 'node:net'
+import http from 'node:http'
 import path from 'node:path'
 import { mkdirSync, mkdtempSync, rmSync, writeFileSync } from 'node:fs'
 import { tmpdir } from 'node:os'
@@ -20,6 +21,7 @@ import {
   reserveServerPort,
   resolveHostTriple,
   spawnSidecar,
+  waitForServer,
   windowsPowerShellOverride,
   writeLastServerPort,
   type SidecarChild,
@@ -29,6 +31,24 @@ function fakeChild(pid = 4321) {
   return { pid, kill: vi.fn() } as unknown as SidecarChild & { kill: ReturnType<typeof vi.fn> }
 }
 
+function listen(server: http.Server, host = '127.0.0.1'): Promise<number> {
+  return new Promise((resolve, reject) => {
+    server.once('error', reject)
+    server.listen(0, host, () => {
+      const address = server.address()
+      if (!address || typeof address === 'string') {
+        reject(new Error('Could not resolve HTTP test port'))
+        return
+      }
+      resolve(address.port)
+    })
+  })
+}
+
+function close(server: http.Server): Promise<void> {
+  return new Promise(resolve => server.close(() => resolve()))
+}
+
 describe('Electron sidecar manager', () => {
   it('maps host platform to existing sidecar target triples', () => {
     expect(resolveHostTriple('darwin', 'arm64')).toBe('aarch64-apple-darwin')
@@ -262,4 +282,20 @@ describe('Electron sidecar manager', () => {
     // Invalid entries are skipped without throwing.
     await expect(reserveServerPort('127.0.0.1', [0, -1, 1.5, 70000])).resolves.toBeGreaterThan(0)
   })
+
+  it('does not treat a raw TCP accept as server readiness without healthy /health', async () => {
+    const server = http.createServer((_request, response) => {
+      response.writeHead(503, { 'content-type': 'application/json' })
+      response.end(JSON.stringify({ status: 'starting' }))
+    })
+    const port = await listen(server)
+
+    try {
+      await expect(waitForServer('127.0.0.1', port, 300)).rejects.toThrow(
+        /desktop server did not report healthy at http:\/\/127\.0\.0\.1:\d+\/health/,
+      )
+    } finally {
+      await close(server)
+    }
+  })
 })

desktop/electron/services/sidecarManager.ts

@@ -164,26 +164,45 @@ export function preferredServerPorts(env: NodeJS.ProcessEnv = process.env): numb
 
 export async function waitForServer(host: string, port: number, timeoutMs = 10_000): Promise<void> {
   const deadline = Date.now() + timeoutMs
+  const healthUrl = `http://${host}:${port}/health`
+  let lastError: Error | null = null
+
   while (Date.now() < deadline) {
-    if (await canConnect(host, port)) return
+    try {
+      await assertServerHealth(healthUrl, Math.min(500, Math.max(100, deadline - Date.now())))
+      return
+    } catch (error) {
+      lastError = error instanceof Error ? error : new Error(String(error))
+    }
     await sleep(150)
   }
-  throw new Error(`desktop server did not start listening on ${host}:${port} within ${Math.round(timeoutMs / 1000)} seconds`)
+
+  const reason = lastError ? `: ${lastError.message}` : ''
+  throw new Error(`desktop server did not report healthy at ${healthUrl} within ${Math.round(timeoutMs / 1000)} seconds${reason}`)
 }
 
-function canConnect(host: string, port: number): Promise<boolean> {
-  return new Promise(resolve => {
-    const socket = net.connect({ host, port, timeout: 200 })
-    socket.once('connect', () => {
-      socket.destroy()
-      resolve(true)
+async function assertServerHealth(healthUrl: string, timeoutMs: number): Promise<void> {
+  const controller = new AbortController()
+  const timeout = setTimeout(() => controller.abort(), timeoutMs)
+  try {
+    const response = await fetch(healthUrl, {
+      cache: 'no-store',
+      signal: controller.signal,
     })
-    socket.once('timeout', () => {
-      socket.destroy()
-      resolve(false)
-    })
-    socket.once('error', () => resolve(false))
-  })
+    if (!response.ok) throw new Error(`healthcheck returned ${response.status}`)
+
+    const contentType = response.headers.get('content-type') ?? ''
+    if (!contentType.toLowerCase().includes('application/json')) {
+      throw new Error(`healthcheck returned non-JSON response from ${healthUrl}`)
+    }
+
+    const body = await response.json().catch(() => null)
+    if (!body || typeof body !== 'object' || !('status' in body) || body.status !== 'ok') {
+      throw new Error(`healthcheck returned invalid response from ${healthUrl}`)
+    }
+  } finally {
+    clearTimeout(timeout)
+  }
 }
 
 function sleep(ms: number): Promise<void> {

desktop/src/components/chat/ChatInput.test.tsx

@@ -1,5 +1,5 @@
 import { fireEvent, render, screen, waitFor } from '@testing-library/react'
-import { beforeEach, describe, expect, it, vi } from 'vitest'
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest'
 import '@testing-library/jest-dom'
 import { act } from 'react'
 
@@ -208,6 +208,10 @@ describe('ChatInput file mentions', () => {
     mocks.listAgents.mockResolvedValue({ activeAgents: [], allAgents: [] })
   })
 
+  afterEach(() => {
+    vi.unstubAllGlobals()
+  })
+
   it('keeps unsent composer drafts isolated when switching between session tabs', async () => {
     const historySessionId = 'history-session'
     useTabStore.setState({
@@ -1158,6 +1162,57 @@ describe('ChatInput file mentions', () => {
     })
   })
 
+  it('ignores pasted images that finish loading after the prompt was sent', async () => {
+    class DeferredFileReader {
+      result: string | ArrayBuffer | null = null
+      onload: ((event: ProgressEvent<FileReader>) => void) | null = null
+
+      readAsDataURL(file: Blob) {
+        pendingReaders.push({ reader: this, file })
+      }
+    }
+    const pendingReaders: Array<{ reader: DeferredFileReader; file: Blob }> = []
+    vi.stubGlobal('FileReader', DeferredFileReader)
+
+    render(<ChatInput compact />)
+
+    const input = screen.getByRole('textbox') as HTMLTextAreaElement
+    const file = new File(['image'], 'late.png', { type: 'image/png' })
+
+    fireEvent.paste(input, {
+      clipboardData: {
+        items: [{
+          type: 'image/png',
+          getAsFile: () => file,
+        }],
+      },
+    })
+    expect(pendingReaders).toHaveLength(1)
+
+    fireEvent.change(input, {
+      target: {
+        value: 'send now',
+        selectionStart: 'send now'.length,
+      },
+    })
+    fireEvent.keyDown(input, { key: 'Enter' })
+
+    expect(mocks.wsSend).toHaveBeenCalledWith(sessionId, {
+      type: 'user_message',
+      content: 'send now',
+      attachments: [],
+    })
+
+    act(() => {
+      pendingReaders[0]!.reader.result = 'data:image/png;base64,LATE'
+      pendingReaders[0]!.reader.onload?.({} as ProgressEvent<FileReader>)
+    })
+
+    await waitFor(() => {
+      expect(screen.queryByAltText(/pasted-image-/)).not.toBeInTheDocument()
+    })
+  })
+
   it('keeps slash and @ popovers outside the drop target clipping context', async () => {
     mocks.search.mockResolvedValueOnce({
       currentPath: '/repo',

desktop/src/components/chat/ChatInput.tsx

@@ -116,6 +116,7 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
   const previousActiveTabIdRef = useRef<string | null>(null)
   const inputRef = useRef(input)
   const attachmentsRef = useRef(attachments)
+  const pasteGenerationRef = useRef(0)
   const setComposerInput = useCallback((value: string) => {
     inputRef.current = value
     setInput(value)
@@ -176,6 +177,9 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
     }
     chatStore.setComposerDraft(sessionId, draft)
   }, [])
+  const invalidatePendingPastes = useCallback(() => {
+    pasteGenerationRef.current += 1
+  }, [])
 
   const isMemberSession = !!memberInfo
   const isActive = chatState !== 'idle'
@@ -222,6 +226,7 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
     }
 
     const nextDraft = activeTabId ? useChatStore.getState().sessions[activeTabId]?.composerDraft : undefined
+    invalidatePendingPastes()
     setComposerInput(nextDraft?.input ?? '')
     setComposerAttachments(nextDraft?.attachments ?? [])
     setPlusMenuOpen(false)
@@ -234,7 +239,7 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
     setEditingQueuedMessageId(null)
     setEditingQueuedMessageText('')
     previousActiveTabIdRef.current = activeTabId
-  }, [activeTabId, saveComposerDraft, setComposerAttachments, setComposerInput])
+  }, [activeTabId, invalidatePendingPastes, saveComposerDraft, setComposerAttachments, setComposerInput])
 
   useEffect(() => {
     return () => {
@@ -711,6 +716,7 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
         displayAttachments: visibleAttachmentPayload,
       })
     }
+    invalidatePendingPastes()
     setComposerInput('')
     setComposerAttachments([])
     useChatStore.getState().clearComposerDraft(activeTabId!)
@@ -817,8 +823,12 @@ export function ChatInput({ variant = 'default', compact = false }: ChatInputPro
       if (!file) continue
 
       const id = `att-${Date.now()}-${Math.random().toString(36).slice(2)}`
+      const pasteGeneration = pasteGenerationRef.current
+      const pastedSessionId = activeTabId
       const reader = new FileReader()
       reader.onload = () => {
+        if (pasteGeneration !== pasteGenerationRef.current) return
+        if (pastedSessionId !== useTabStore.getState().activeTabId) return
         setComposerAttachments((prev) => [
           ...prev,
           {

desktop/src/components/layout/Sidebar.test.tsx

@@ -322,6 +322,19 @@ describe('Sidebar', () => {
     await waitFor(() => expect(fetchSessions).toHaveBeenCalled())
   })
 
+  it('exposes the full session title as a row tooltip when the label is truncated', () => {
+    const longTitle = '这是一个非常非常长的会话标题，用来验证侧边栏截断后仍然可以通过气泡查看完整内容'
+    useSessionStore.setState({
+      sessions: [
+        makeSession('session-long-title', longTitle, '/workspace/alpha', '2026-05-15T10:00:00.000Z'),
+      ],
+    })
+
+    render(<Sidebar />)
+
+    expect(screen.getByRole('button', { name: new RegExp(longTitle) })).toHaveAttribute('title', longTitle)
+  })
+
   it('reorders project groups by dragging project headers while preserving expanded state', async () => {
     const base = new Date('2026-05-15T10:00:00.000Z').getTime()
     useSessionStore.setState({

desktop/src/components/layout/Sidebar.tsx

@@ -975,6 +975,7 @@ export function Sidebar({ isMobile = false, onRequestClose }: SidebarProps) {
                                     }
                                   `}
                                   aria-pressed={isBatchMode ? selectedSessionIds.has(session.id) : undefined}
+                                  title={session.title || 'Untitled'}
                                 >
                                   <span className="flex min-w-0 items-center gap-2">
                                     {isBatchMode ? (

src/server/__tests__/provider-presets.test.ts

@@ -141,6 +141,7 @@ describe('provider presets API', () => {
     expect(zhipu?.apiKeyUrl).toBe('https://www.bigmodel.cn/invite?icode=d41B2qi8Z5xNwTGLNPPF3OZLO2QH3C0EBTSr%2BArzMw4%3D')
     expect(zhipu?.promoText).toContain('cc-haha')
     expect(zhipu?.defaultEnv?.CC_HAHA_SEND_DISABLED_THINKING).toBe('1')
+    expect(zhipu?.modelContextWindows?.['glm-5.2']).toBe(200000)
     expect(zhipu?.modelContextWindows?.['glm-5.1']).toBe(200000)
     expect(zhipu?.modelContextWindows?.['glm-4.5-air']).toBe(128000)
     expect(kimi?.apiKeyUrl).toBe('https://platform.kimi.com/console/api-keys')

src/server/__tests__/proxy-transform.test.ts

@@ -63,6 +63,34 @@ describe('anthropicToOpenaiChat', () => {
     expect(result.stop).toEqual(['END', 'STOP'])
   })
 
+  test('omits Anthropic sampling params by default for OpenAI-compatible providers', () => {
+    const req: AnthropicRequest = {
+      model: 'glm-5.2',
+      max_tokens: 100,
+      temperature: 0.7,
+      top_p: 0.9,
+      messages: [{ role: 'user', content: 'Hi' }],
+    }
+
+    const result = anthropicToOpenaiChat(req)
+    expect(result.temperature).toBeUndefined()
+    expect(result.top_p).toBeUndefined()
+  })
+
+  test('can explicitly pass sampling params for chat providers that accept them', () => {
+    const req: AnthropicRequest = {
+      model: 'gpt-4',
+      max_tokens: 100,
+      temperature: 0.7,
+      top_p: 0.9,
+      messages: [{ role: 'user', content: 'Hi' }],
+    }
+
+    const result = anthropicToOpenaiChat(req, { passSamplingParams: true })
+    expect(result.temperature).toBe(0.7)
+    expect(result.top_p).toBe(0.9)
+  })
+
   test('tools conversion', () => {
     const req: AnthropicRequest = {
       model: 'gpt-4',
@@ -407,6 +435,34 @@ describe('anthropicToOpenaiResponses', () => {
     expect(result.input).toEqual([{ type: 'message', role: 'user', content: 'Hello' }])
   })
 
+  test('omits Anthropic sampling params by default for Responses-compatible providers', () => {
+    const req: AnthropicRequest = {
+      model: 'glm-5.2',
+      max_tokens: 100,
+      temperature: 0.7,
+      top_p: 0.9,
+      messages: [{ role: 'user', content: 'Hi' }],
+    }
+
+    const result = anthropicToOpenaiResponses(req)
+    expect(result.temperature).toBeUndefined()
+    expect(result.top_p).toBeUndefined()
+  })
+
+  test('can explicitly pass sampling params for Responses providers that accept them', () => {
+    const req: AnthropicRequest = {
+      model: 'gpt-4o',
+      max_tokens: 100,
+      temperature: 0.7,
+      top_p: 0.9,
+      messages: [{ role: 'user', content: 'Hi' }],
+    }
+
+    const result = anthropicToOpenaiResponses(req, { passSamplingParams: true })
+    expect(result.temperature).toBe(0.7)
+    expect(result.top_p).toBe(0.9)
+  })
+
   test('tools conversion uses top-level name', () => {
     const req: AnthropicRequest = {
       model: 'gpt-4o',

src/server/config/providerPresets.json

@@ -60,6 +60,7 @@
       "CC_HAHA_SEND_DISABLED_THINKING": "1"
     },
     "modelContextWindows": {
+      "glm-5.2": 200000,
       "glm-5.1": 200000,
       "glm-5-turbo": 200000,
       "glm-4.5-air": 128000

src/server/proxy/transform/anthropicToOpenaiChat.ts

@@ -21,6 +21,7 @@ type OpenAIChatImageContentMode = 'vision' | 'text_only'
 type OpenAIChatTransformOptions = {
   roundTripReasoningContent?: boolean
   passThinkingToggle?: boolean
+  passSamplingParams?: boolean
   imageContentMode?: OpenAIChatImageContentMode
 }
 
@@ -68,9 +69,12 @@ export function anthropicToOpenaiChat(
   // Claude Code sends very large values (e.g. 128K) that exceed many
   // providers' limits (DeepSeek: 8192, etc.).
 
-  // temperature & top_p
-  if (body.temperature !== undefined) result.temperature = body.temperature
-  if (body.top_p !== undefined) result.top_p = body.top_p
+  // Claude Code sends Anthropic sampling params that some compatible
+  // providers reject. Keep them opt-in for providers known to accept them.
+  if (options.passSamplingParams) {
+    if (body.temperature !== undefined) result.temperature = body.temperature
+    if (body.top_p !== undefined) result.top_p = body.top_p
+  }
 
   // stop_sequences → stop
   if (body.stop_sequences && body.stop_sequences.length > 0) {

src/server/proxy/transform/anthropicToOpenaiResponses.ts

@@ -17,6 +17,7 @@ import { stripLeadingBillingHeader } from './billingHeader.js'
 export type OpenAIResponsesTransformOptions = {
   /** Stable cache routing key, forwarded as `prompt_cache_key`. */
   cacheKey?: string
+  passSamplingParams?: boolean
 }
 
 /**
@@ -59,9 +60,12 @@ export function anthropicToOpenaiResponses(
   // max_tokens — omit to let upstream provider use its own default/max.
   // Claude Code sends very large values that exceed many providers' limits.
 
-  // temperature & top_p
-  if (body.temperature !== undefined) result.temperature = body.temperature
-  if (body.top_p !== undefined) result.top_p = body.top_p
+  // Claude Code sends Anthropic sampling params that some compatible
+  // providers reject. Keep them opt-in for providers known to accept them.
+  if (options.passSamplingParams) {
+    if (body.temperature !== undefined) result.temperature = body.temperature
+    if (body.top_p !== undefined) result.top_p = body.top_p
+  }
 
   // tools
   if (body.tools && body.tools.length > 0) {

src/services/compact/autoCompact.test.ts

@@ -32,6 +32,7 @@ describe('model context window resolution', () => {
     expect(getContextWindowForModel('deepseek-v4-pro')).toBe(1_000_000)
     expect(getContextWindowForModel('MiniMax-M2.7')).toBe(204_800)
     expect(getContextWindowForModel('kimi-k2.6')).toBe(262_144)
+    expect(getContextWindowForModel('zai-org/GLM-5.2')).toBe(200_000)
     expect(getContextWindowForModel('glm-5.1')).toBe(200_000)
     expect(getContextWindowForModel('glm-4.5-air')).toBe(128_000)
   })
@@ -61,6 +62,7 @@ describe('model context window resolution', () => {
 
   test('derives auto-compact thresholds from provider context windows', () => {
     expect(getAutoCompactThreshold('deepseek-v4-pro')).toBe(967_000)
+    expect(getAutoCompactThreshold('zai-org/GLM-5.2')).toBe(167_000)
     expect(getAutoCompactThreshold('glm-5.1')).toBe(167_000)
     expect(getAutoCompactThreshold('glm-4.5-air')).toBe(95_000)
     expect(getAutoCompactThreshold('kimi-k2.6')).toBe(229_144)

src/utils/model/modelContextWindows.ts

@@ -19,6 +19,7 @@ const DIRECT_MODEL_CONTEXT_WINDOWS: Record<string, number> = {
   'minimax-m3': 1_000_000,
   'minimax-m2.7': 204_800,
   'minimax-m2.7-highspeed': 204_800,
+  'glm-5.2': 200_000,
   'glm-5.1': 200_000,
   'glm-5': 200_000,
   'glm-5-turbo': 200_000,
@@ -36,6 +37,7 @@ const PATTERN_MODEL_CONTEXT_WINDOWS: Array<[RegExp, number]> = [
   [/^openai\/gpt-5(?:[.-]\d+)?\b/i, 400_000],
   [/^google\/gemini-(?:2\.0|2\.5|3)/i, 1_048_576],
   [/^gemini-(?:2\.0|2\.5|3)/i, 1_048_576],
+  [/^zai-org\/glm-5\.2\b/i, 200_000],
   [/^qwen\/qwen3\.5-(?:plus|flash)\b/i, 1_000_000],
   [/^qwen3\.5-(?:plus|flash)\b/i, 1_000_000],
   [/^qwen\/qwen3-max\b/i, 262_144],