fix: prevent transient scheduled task write failures

Cron task metadata writes can overlap closely enough that a timestamp-only
$temp filename is not stable. Harden the atomic write path by using a
collision-resistant temp name and retrying once when rename reports ENOENT.

Constraint: Scheduled task persistence must keep atomic replace semantics in ~/.claude/scheduled_tasks.json
Rejected: Add explicit file locking around every scheduled task write | more coordination overhead than the observed transient rename race warrants
Confidence: high
Scope-risk: narrow
Directive: Keep scheduled task writes collision-resistant and retriable; do not revert to timestamp-only temp file names without reproducing concurrent writes
Tested: bun test src/server/__tests__/scheduled-tasks.test.ts
Not-tested: bun test src/server/__tests__/cron-scheduler.test.ts (existing timeouts and unrelated assertion failures in this checkout)

src/server/__tests__/scheduled-tasks.test.ts

@@ -2,7 +2,7 @@
  * Unit tests for CronService, SearchService, and Scheduled Tasks API
  */
 
-import { describe, it, expect, beforeEach, afterEach } from 'bun:test'
+import { describe, it, expect, beforeEach, afterEach, spyOn } from 'bun:test'
 import * as fs from 'fs/promises'
 import * as path from 'path'
 import * as os from 'os'
@@ -139,6 +139,41 @@ describe('CronService', () => {
       service.createTask({ cron: '* * * * *', prompt: '' }),
     ).rejects.toThrow()
   })
+
+  it('should retry the atomic write when rename returns ENOENT', async () => {
+    const originalRename = fs.rename
+    let renameCalls = 0
+
+    const renameSpy = spyOn(fs, 'rename')
+    renameSpy.mockImplementation(async (...args) => {
+      renameCalls += 1
+
+      if (renameCalls === 1) {
+        const error = new Error(
+          'ENOENT: no such file or directory, rename tmp -> scheduled_tasks.json',
+        ) as NodeJS.ErrnoException
+        error.code = 'ENOENT'
+        throw error
+      }
+
+      return originalRename(...args)
+    })
+
+    try {
+      const task = await service.createTask({
+        cron: '0 9 * * *',
+        prompt: 'Retry rename once',
+      })
+
+      const tasks = await service.listTasks()
+      expect(task.id).toBeDefined()
+      expect(tasks).toHaveLength(1)
+      expect(tasks[0]?.prompt).toBe('Retry rename once')
+      expect(renameCalls).toBe(2)
+    } finally {
+      renameSpy.mockRestore()
+    }
+  })
 })
 
 // ─── SearchService tests ────────────────────────────────────────────────────

src/server/services/cronService.ts

@@ -32,6 +32,8 @@ type TasksFile = {
   tasks: CronTask[]
 }
 
+const TASKS_FILE_WRITE_ATTEMPTS = 2
+
 export class CronService {
   /** 任务文件路径 */
   private getTasksFilePath(): string {
@@ -134,21 +136,32 @@ export class CronService {
   private async writeTasksFile(data: TasksFile): Promise<void> {
     const filePath = this.getTasksFilePath()
     const dir = path.dirname(filePath)
-    await fs.mkdir(dir, { recursive: true })
+    const contents = JSON.stringify(data, null, 2) + '\n'
+    let lastError: Error | undefined
 
-    const tmpFile = `${filePath}.tmp.${Date.now()}`
-    try {
-      await fs.writeFile(
-        tmpFile,
-        JSON.stringify(data, null, 2) + '\n',
-        'utf-8',
-      )
-      await fs.rename(tmpFile, filePath)
-    } catch (err) {
-      await fs.unlink(tmpFile).catch(() => {})
-      throw ApiError.internal(
-        `Failed to write scheduled tasks: ${(err as Error).message}`,
-      )
+    for (let attempt = 0; attempt < TASKS_FILE_WRITE_ATTEMPTS; attempt++) {
+      const tmpFile = `${filePath}.tmp.${process.pid}.${Date.now()}.${crypto.randomBytes(6).toString('hex')}`
+
+      try {
+        await fs.mkdir(dir, { recursive: true })
+        await fs.writeFile(tmpFile, contents, 'utf-8')
+        await fs.rename(tmpFile, filePath)
+        return
+      } catch (err) {
+        lastError = err as Error
+        await fs.unlink(tmpFile).catch(() => {})
+
+        if (
+          (err as NodeJS.ErrnoException).code !== 'ENOENT' ||
+          attempt === TASKS_FILE_WRITE_ATTEMPTS - 1
+        ) {
+          break
+        }
+      }
     }
+
+    throw ApiError.internal(
+      `Failed to write scheduled tasks: ${lastError?.message ?? 'unknown error'}`,
+    )
   }
 }