openclaw - 💡(How to fix) Fix [Bug]: managed macOS update handoff can fail with uv_cwd after LaunchAgent stop

Bug type

Crash / restart-handoff failure

Beta release blocker

No

Summary

On macOS with a LaunchAgent-managed gateway and npm-installed OpenClaw, a managed update can stop the gateway and then fail the restart handoff because the spawned openclaw update --yes process crashes with Error: ENOENT: no such file or directory, uv_cwd.

The visible result is a managed-service-handoff-failed / Gateway restart update error (npm) notification even though openclaw update status may later report the install is already up to date.

This looks related to the general uv_cwd bug in #73676, but here it breaks the managed update/restart path itself rather than only an interactive CLI command.

Steps to reproduce

1. Run OpenClaw on macOS as a LaunchAgent-managed gateway.
2. Install OpenClaw globally via npm/pnpm-style CLI install.
3. Trigger a managed update/restart flow that runs openclaw update --yes from the gateway side.
4. Observe the restart handoff fail after the LaunchAgent is stopped.

I do not yet have a minimal standalone repro for the exact cwd state inside the managed update helper, but the production handoff log below shows the failure clearly.

Expected behavior

Managed update should either:

• sanitize/fallback the working directory before invoking the updater, or
• refuse the handoff safely before stopping the LaunchAgent.

In either case, it should not stop the service and then crash on uv_cwd.

Actual behavior

The managed update helper starts openclaw update --yes, stops the LaunchAgent, and then the updater crashes immediately with:

Error: ENOENT: no such file or directory, uv_cwd

The user receives a restart/update error notification:

Gateway restart update error (npm)
Reason: managed-service-handoff-failed

At the same time, openclaw update status can report:

Install  pnpm
Channel  stable (default)
Update   pnpm · up to date · npm latest 2026.5.18

OpenClaw version

2026.5.18

Operating system

macOS

Install method

npm / pnpm-managed global install

Model

NOT_RELEVANT

Provider / routing chain

NOT_RELEVANT

Additional provider/model setup details

NOT_RELEVANT

Logs, screenshots, and evidence

Managed handoff log:

[2026-05-18T23:28:59.261Z] starting managed update command: openclaw update --yes
[2026-05-18T23:28:59.262Z] managed update command pid=51881
Stopped LaunchAgent: gui/501/ai.openclaw.gateway
node:internal/bootstrap/switches/does_own_process_state:142
    cachedCwd = rawMethods.cwd();
                           ^

Error: ENOENT: no such file or directory, uv_cwd
    at process.wrappedCwd [as cwd] (node:internal/bootstrap/switches/does_own_process_state:142:28)
    at isMainModule (file:///opt/homebrew/lib/node_modules/openclaw/dist/is-main-myFrAEl9.js:13:92)
    at file:///opt/homebrew/lib/node_modules/openclaw/dist/index.js:38:16
    at ModuleJob.run (node:internal/modules/esm/module_job:343:25)
    at async onImport.tracePromise.__proto__ (node:internal/modules/esm/loader:665:26)
    at async asyncRunEntryPointWithESMLoader (node:internal/modules/run_main:117:5) {
  errno: -2,
  code: 'ENOENT',
  syscall: 'uv_cwd'
}

Node.js v22.22.2
[2026-05-18T23:29:18.476Z] managed update command exited code=1 signal=null

Persistent handoff marker left behind after the failure:

{"kind":"gateway-supervisor-restart-handoff","version":1,"intentId":"5ddc42fb-81c4-4c66-84fa-084cefbb8b0f","pid":51435,"processInstanceId":"08729a78-0e23-40c9-b62e-dfaf6ce76f38","createdAt":1779146937662,"expiresAt":1779146997662,"reason":"update.run","source":"gateway-update","restartKind":"update-process","supervisorMode":"launchd"}

Impact and severity

Affected: macOS LaunchAgent-managed gateways using managed self-update Severity: High for operator trust / availability Frequency: Seen in production during update Consequence: The update channel reports a failed managed handoff after the service stop, leaving the operator unsure whether the gateway is healthy and whether the update actually succeeded.

Additional information

• openclaw doctor --non-interactive completes without surfacing this handoff failure.
• openclaw doctor --fix --non-interactive also does not clear or explain the failed restart handoff state.
• This seems adjacent to both:
  • #73676 (uv_cwd when cwd disappears)
  • #81894 / #52313 (managed/self-update restart reliability)

Suggested fix

1. In the managed update/restart path, resolve process.cwd() defensively before stopping the LaunchAgent.
2. If cwd lookup fails, fall back to a stable directory such as the user home or package root before invoking the CLI entrypoint.
3. Treat uv_cwd during managed update as a first-class, user-visible recovery case with a concrete repair message.
4. Add an integration test for macOS LaunchAgent managed updates where the initiating process has an invalid cwd.