Title

channels start reports success but bot stays offline after channels stop

Description

Running nemoclaw <sandbox> channels stop telegram correctly disables the Telegram bridge — the bot goes offline as expected.

Running nemoclaw <sandbox> channels start telegram immediately afterwards reports success (e.g. ✓ Marked telegram enabled for '<sandbox>'), but the Telegram bot never comes back online. It does not respond to DMs or group mentions.

Expected: channels start should fully restore the bridge so the bot resumes responding, mirroring the stop behavior.

Reproduction Steps

1. Onboard a sandbox with Telegram configured (nemoclaw <sandbox> channels add telegram with a valid bot token).
2. Confirm the bot responds to a DM.
3. Run nemoclaw <sandbox> channels stop telegram, accept the rebuild prompt. Verify the bot stops responding.
4. Run nemoclaw <sandbox> channels start telegram (same shell session, back-to-back), accept the rebuild prompt.
5. Send a DM to the bot — no response.

Verification — check whether Telegram is wired into the sandbox config

After step 4, inspect openclaw.json inside the sandbox to confirm whether the channel is actually registered:

docker exec -it openshell-cluster-nemoclaw bash
kubectl exec -it <sandbox-pod> -n Openshell -- su - sandbox
cat .openclaw/openclaw.json

Compare the Telegram entry before stop, after stop, and after start.

Notes

• channels stop works as expected; only start after stop exhibits this.
• Suspected area: sandboxChannelsSetEnabled in src/lib/actions/sandbox/policy-channel.ts:634 flips the disabled flag in the registry and calls promptAndRebuild, but the rebuild may not be re-attaching the bridge to the OpenShell gateway the same way channels add does
  (which explicitly calls applyChannelAddToGatewayAndRegistry).

Environment (fill in before submitting)

• OS:
• Node.js: (node --version)
• Docker: (docker version)
• NemoClaw: (nemoclaw --version)

The bug template will auto-apply bug and status: triage labels.