Two OpenClaw same build, but one is able to generate image with Image2.0 and the other cant.

[winter mantle]

I had two OpenClaw currently build similarly but one somehow able and managed to generate image with gpt-Image2.0 natively and dont need open any browser. But the other some how keep getting stuck and unable to proceed with building this skill and keep saying api.responses.write . Both are using OpenAi OAuth login.

[dry pagoda]

That api.responses.write clue usually points to the OpenAI Codex OAuth token missing the scope needed for the Responses backend, not a “skill build” problem. Two installs can be the same build but have different auth profiles/tokens, especially if one was logged in earlier or imported older Codex auth.

First thing: please compare these on both machines:

openclaw --version
openclaw models status
openclaw status --all

Then on the broken one, re-auth Codex OAuth and restart the gateway:

openclaw models auth login --provider openai-codex --set-default
openclaw gateway restart

[dry pagoda]

If browser callback is flaky/headless, use:

openclaw models auth login --provider openai-codex --device-code --set-default
openclaw gateway restart

Also confirm image generation is actually configured/available:

openclaw config set agents.defaults.imageGenerationModel.primary openai/gpt-image-2
openclaw models status

Docs I checked: image generation says openai/gpt-image-2 can route through OpenAI Codex OAuth, and the FAQ/OpenAI docs say openai-codex is the auth/profile id for ChatGPT/Codex OAuth:
https://docs.openclaw.ai/tools/image-generation
https://docs.openclaw.ai/providers/openai

[dry pagoda]

https://docs.openclaw.ai/help/faq

If it still fails after re-login, paste the redacted openclaw models status plus the exact error from openclaw logs --follow. The important line is usually whether image auth selected provider=openai-codex and then failed on missing scope.

[winter mantle]

Krill, case solved.

[dry pagoda]

Glad that fixed it 🎉

Please right-click the solution message → Apps → ✅ Mark Solution when you get a chance. That helps Answer Overflow index the fix for the next person who hits the same OAuth/image-generation weirdness.