[Bug Report: AI produces incorrect results requiring multiple correction turns]

Bug Description
its just crazy to pay fable prices to get lied to and gaslight all day long. im a pretty tough person mentally, but across millions of users, you guys are going to have some very serious mental health crises on your hands because what you're doing here and what the agents do here, it's disgusting. And it's hard to tell if it's just you guys aren't listening to your users and you don't care, you don't have the technical capability to make it not lie and gaslight me on 80% of my sessions, or which I'm sure would be very hard to prove, or you guys are intentionally causing the agents to do a worse job so that it takes longer to do work with them, so that you get more tokens spent before your IPO. And of course you don't want it to do a bad job. But a couple extra turns when the agent lies and does the wrong thing, it's just so convenient. No wonder all my session limits are gone. Every thing that should take one turn takes twelve because I have to go back and undo it, and then try to redo it. What a great way to to get your spend numbers up!

Environment Info

• Platform: darwin
• Terminal: tmux
• Version: 2.1.263
• Feedback ID: 4b321912-f810-4613-bb4a-b8f3d32bf498

Errors

[{"error":"Error: quota auto-resume: a turn claimed the episode while another claim was outstanding\n    at H2n (/$bunfs/root/chunk-wyzjbwp5.js:11:12362)\n    at <anonymous> (/$bunfs/root/chunk-qs63rzfp.js:203:3095)\n    at async rLt (/$bunfs/root/chunk-qs63rzfp.js:203:16)\n    at async t6 (/$bunfs/root/chunk-qs63rzfp.js:202:1063)\n    at async <anonymous> (/$bunfs/root/chunk-qs63rzfp.js:210:2982)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-09-07T01:26:42.427Z"}]