Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore. You should be designing loops that prompt your agents.
@steipete
-
Copilot SDK can now be used as a harness
By
–
That works for quite a while; new is that you can even use Copilot SDK as harness.
-
Using Codex to drive OpenClaw: a correction
By
–
it’s the other way around, you can use codex to drive OpenClaw.
-
GPT 5.5 is larger, doubles usage, no silent quota reduction
By
–
You probably moved to GPT 5.5 which is a far larger model and takes double usage. There's no silent reduction in quotas.
-
Summoning AI help via sag.sh when distracted
By
–
I told codex to use http://
sag.sh whenever I'm distracted and it needs my help to be unblocked, and ever once it a while I hear it talking to me, and it's the coolest thing ever. (e.g. for releases, that needs npm and is 1Password-gated) -
Anthropic restrictive, Codex is far better for code review and allows sub.
By
–
Hmmm Anthropic is very restrictive, claude -p no longer works so no. Codex is far better at code review tho and allows using the sub!
-
Codex works unprompted without frameworks or third-party
By
–
The funny part is nothing of that is needed anymore – codex just does this unprompted on its own, no framework or 3rd-party needed.
-
Teaching Codex to automate QA testing of OpenClaw via VNC and browser
By
–
Been teaching codex to be my QA assistant. For every commit it creates a user-test scenario and uses webVNC (crabbox), computer/browser use (peekaboo/mcporter) to test OpenClaw like a user/QA person would. This runs in the background and opens PRs with fixes.
-
Codex finds most issues; other models offer light review
By
–
nah, but codex finds by far the most issues. You can get a light review with other models