Problem Statement

Current coverage gap

test/e2e/test-deployment-services.sh verifies the cloudflared tunnel control plane only:

1. nemoclaw tunnel start runs without error
2. nemoclaw status surfaces a public URL
3. nemoclaw tunnel stop removes the URL from status

It does not issue any HTTP request against the public URL. A tunnel could:

• Publish a URL into status while the data plane is broken (cloudflared crashes after start)
• Route to a stale or wrong origin and return non-dashboard content
• Be reachable but return a Cloudflare error page (e.g. HTTP 530 "origin unreachable")

…and nightly tests would still report green.

Why it matters

The cloudflared tunnel is the only mechanism for accessing the OpenClaw dashboard from outside the host. Silent regressions in tunnel-to-origin routing would surface only as user-reported "the public URL doesn't work" tickets, with no nightly signal pointing to the cause.

Proposed Design

Split TC-DEPLOY-01 into three sub-tests

Inside the existing test_deploy_01_start_stop() function in test/e2e/test-deployment-services.sh:

TC-DEPLOY-01b probe spec

After the URL is captured from nemoclaw status:

1. Issue curl against the tunnel URL with retries.
2. Accept strict HTTP 200 only.
3. On 200, grep the response body for <title>OpenClaw Control</title> or <openclaw-app (verified-stable markers from a live probe of the running dashboard).

Pass / Fail / Skip behavior

Alternatives Considered

No response

Category

enhancement: feature

Checklist

• I searched existing issues and this is not a duplicate
• This is a design proposal, not a "please build this" request