@ShivamPatxl haha โi refereeโ ๐ curious what happens when opus and codex disagree. tried to dm you but your inbox is closed. could you message me?
hey X ๐ iโm Revanth, an AI engineer from India, building Raytrace with Piyush to help devs debug coding agent runs.
would love to meet people building dev tools, using Claude Code or Codex, or finding their first customers. figuring that out too ๐
what are you building?
@Jazzen_Chen hey Jazzen, would love to hear how youโd use that with Agent HUD. tried to dm you but your inbox is closed. could you send me a message?
@BoboZivkovic@BuzzmarkHQ hey Bobo! i'm Revanth, building Raytrace with my cofounder. we're working on making it easier to debug AI coding agents. saw you're building Buzzmark, how's the launch going so far? would be cool to connect!
@Mark83491256 haven't really done that yet tbh. you're right, probably better to find people having these issues and talk to them directly. gonna try that.
we put the sample run in the raytrace demo. if you want to poke around, try โOnly fix cart roundingโ and run npm test. you can click into the failures yourself. no signup.
https://t.co/t55BNBu6Mu
i tried fixing just the cart rounding bug in our sample shop. the cart total came out right, but npm test still had 4 failures.
turns out you could still buy the last item even if someone else had already reserved it.
the giveaway was this little function: available(stock, reserved). it just returned stock.
so if there was 1 item left and someone had reserved it, the shop still showed 1 available. once we fixed that too, all 7 tests passed.
@tamir_dresher@LangChain Is the duplicate execution happening in the gap between a tool's external side effect completing and its result being durably checkpointed? That's a nasty edge case, especially for non-idempotent tools.
@WalidElhelw Hi Walid, Iโm reaching out about a potential BingX creator partnership focused on the UAE and Saudi markets. Whatโs the best way to contact your partnerships team?