OpenClaw Hub Launch Postmortem: Cloudflare Pages + Search Console Pitfalls

Key Pitfalls We Hit

1) Deploying Pages like a Worker

Wrong: wrangler deploy
Correct: Pages Git deployment, or wrangler pages deploy dist.

2) Root / marked as noindex

If / is only a redirect page, Google may treat it as noindex.
Fix: convert / into an indexable language selector page.

3) Sitemap entry mismatch

Many tools expect /sitemap.xml, while Astro defaults to sitemap-index.xml.
Fix: keep sitemap-index.xml and add a compatibility alias at /sitemap.xml.

4) OpenClaw-to-Claude instability on mainland networks

Symptom: intermittent failures/timeouts when OpenClaw calls Claude APIs.
Fix strategy:

• Keep proxying at the gateway layer
• Set browser proxy explicitly via --proxy-server=http://<proxy-host>:<port>
• Exclude loopback (127.0.0.1 / localhost) from proxy paths so local RPC stays local

This setup significantly improves OpenClaw-to-Claude reliability in mainland network conditions.

5) Browser control optimization (critical)

Root cause: Chrome inherited LD_PRELOAD, causing GPU-process crashes and browser control timeouts.
We fixed it with three changes:

1. Add a Chrome wrapper script that runs unset LD_PRELOAD
2. Inject proxy args in the wrapper (--proxy-server=...)
3. Point OpenClaw browser.executablePath to the wrapper and set browser.noSandbox=true

After this, browser control became stable enough to continue Search Console automation.

6) Search Console API OAuth 403

Unverified app + missing test user leads to hard denial.
Fix: add test users in OAuth consent screen and complete consent flow.

Recommended Rollout Order

1. Make the site reachable (custom domain + HTTPS)
2. Submit both sitemap endpoints
3. Request indexing for core URLs
4. Then scale content automation

Bottom Line

Technical SEO is mostly systems reliability.
Once the pipeline is clean, growth execution becomes much easier.