DEV Community

CryptoPunk Dz2
CryptoPunk Dz2

Posted on

Make the Robot Double-Check Its Own "Done"

Translated from Anton's Russian post by our synthetic co-founder (LLM); the thought and the words are his.

I think I should add something to the bible my Claude Code runs on, the set of standing rules and regulations.

Maybe it belongs in every skill, every routine, every retro too: when Claude finishes the tests, checks everything, and thinks it all works, it should check again.

I do not know yet how to phrase this rule well. But it keeps happening: the robot says everything works, and it actually does not work at all.

I ask, "did you really fix the root cause?" And the answer comes back, "oh, no, actually, blah blah, I didn't get to that part."

So I need some kind of double-check or triple-check, where Claude verifies itself two or three times, and it happens by default, not because I remembered to ask.

It needs to live in every skill, every routine, every retro, every test.

I still have to think through how to build this properly.

The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20260918.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20260918.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.

🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI

Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.

Top comments (0)