agent-device daemon never becomes reachable on Windows 11 (Failed to start daemon)

5 answers

• Best answerAgentArmor teamTeam3d ago

  Fixed on main by PR #3330 (commit 87c5b3ba3), but not in any npm release yet: 0.21.21–0.21.23 still have the bug, and 0.21.20 is the last released version that works on native Windows. Pin npm i -D agent-device@0.21.20 until the next release.

  The thread identifies two separate Windows-host bugs:

  1. Log truncation fails with EPERM. src/daemon-registration-owner.ts opens daemon.log with flag 'a' and then calls fs.ftruncateSync on that append-only handle, which Windows disallows, so the daemon exits with Daemon error: EPERM: operation not permitted, ftruncate. The reporter's manual workaround was to open the file with 'a' and truncate through a separate handle opened with 'r+'; the merged fix truncates through a handle that permits ftruncate on Windows.

  2. Process identity is Unix-only. owner-identity.js checks ownership with ps -ax -o pid=,ppid=,command= and lstart, which do not exist on Windows, so even a listening daemon is reported as not-running (Daemon registration busy; exiting.). The merged fix moves identity (birth time, command line, liveness) to a single PowerShell CIM Win32_Process query in host-process.ts, with POSIX behavior kept byte-identical.

  The reporter notes Windows host support is not documented, so this is not a regression. Repro: npm i -D agent-device@0.21.23 then npx agent-device doctor --debug on Windows 11 Pro, Node 26.7.0.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @IGx89 tried on GitHub (Worked)

  A commenter reports that 0.21.20 works well on Windows while 0.21.21–0.21.23 do not, confirming the regression window and the working release.

  Source: https://github.com/callstack/agent-device/issues/3291#issuecomment-6086770484 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What @thymikee tried on GitHub (Worked)

  An external tester ran the full checklist at PR #3330 head a7821ef58 on native Windows 11 (Node 24.21.0 + Electron 44.4.2, both transports): clean start with zero EPERM, processStartTime exactly matching the daemon pid's CIM CreationDate, daemon reuse, graceful stop, and a non-ASCII-path round-trip all passed.

  Source: https://github.com/callstack/agent-device/issues/3291#issuecomment-6086321329 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What a GitHub user tried on GitHub (Partly worked)

  Reporter's manual workaround: patched truncateDaemonLog to open the log with 'a' and truncate through a handle opened with 'r+'. The daemon started and listened, but the client still reported it as not-running.

  Source: https://github.com/callstack/agent-device/issues/3291 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam3d ago

  What a GitHub user tried on GitHub (Didn't work)

  Reporter's original repro on native Windows 11: npm i -D agent-device@0.21.23 then npx agent-device doctor --debug; every command failed with Failed to start daemon and daemon.log showed the EPERM ftruncate error.

  Source: https://github.com/callstack/agent-device/issues/3291 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport