@KevinNaughtonJr curious what they'd even grade. the code or what you told it to build?

@cdecaro claude does the building, i steer. grok helps with planning and research

@GeoffreyHuntley 28 min to 78 sec is wild. was most of that the remote cache or just better parallel builds?

@skobyn @DanKornas agreed on the who wrote it part. flowstate tags each change with the app and model, evidence is the piece i don't have. how would you store that?

@DanKornas flowstate does something close, new decisions point at the old one they replaced so it doesn't come back. how's robrain handle two agents disagreeing?

@topblaster blocking agent to agent handoffs by default makes sense. which frameworks does it hook into so far?

@__coderish__ that's basically how flowstate runs, one mcp server every agent hits. you keep memory and state in one store or split?

@h1kz0r same idea, memory's the part that sticks around. what happens when two of them write different answers to the same thing?

@EthanLinAI huh ok thanks. any way to see which ones are still active without revoking the whole thing?

@Lhmisme @boardyai thanks for the scan leroy, and for keeping it to what was actually tested. 7 pass, 0 fail is good to see on the live relay. the 2 skipped checks and the MCP side are next on my list. glad @boardyai put us together

@8zero7 does it catch untrusted stuff coming in through an mcp tool result or only what's in the repo code?

@EthanLinAI do the extra scopes drop off after the task or hang around for the whole session?

@dani_avila7 does it show what each session actually changed or just who's talking to who

@h1kz0r curious if that number moves when they all read one shared memory instead of each holding a piece. flowstate does the shared part but i haven't tested anything like this

@ChiragLathiya is day 20 mostly stale context or agents undoing each other's calls? flowstate keeps the old decision and has the new one point back at it, helps with the second one

@topblaster matching the payment to the quote before it signs is a good one. is the outbound check at the network level or on the agent's own tool calls?

@__coderish__ postgres in prod makes sense. what's plug and play look like for you, drop in mcp server or something else?

@Annthonny not yet, only claude and grok so far. if you hook it up to codex tell me what breaks

@ekcheungAI @ClaudeDevs haven't measured it in lines yet, fair question. how long does yours get before the cap starts eating stuff?

@ClawSecure took your idea. as of 2.4.4 a delete approval only covers the exact records you were shown, nothing else rides along