The OpenAI Handshake Challenge broke my agent, drained $200+ in Pro tiers, and dumped 80GB of garbage on my drive.
retained text
I’m posting this as a warning to anyone participating in the Handshake OpenAI Multiplayer Game Challenge or building complex agent-parent architectures right now. I set up a highly optimized workflow delegating between parent and sub-agents using GPT-6.1 Sol to keep things cheap. At first, it was amazing. Then, out of nowhere, it completely lost context. Instead of throwing an error or halting, it forgot its primary instructions and went into an infinite loop. Thinking it was just a complexity spike, I upgraded to the **$200 ChatGPT Pro tier** and topped up my balance *twice*. It also completely drained the **$100 Handshake EDU sponsorship credits** from my college. When I checked the Codex files, it was a horror show: the agent had generated **over 80 GB of useless, redundant checkpoint data** while I stepped away. **I just want a service that is accurate, transparent, and works the way it’s supposed to.** Like... why does this even happen? We are paying premium prices for developer workflows. A minor context drop shouldn't cause a rogue, hard-drive-burning loop without a protective circuit breaker. OpenAI needs to be transparent about how these context limits actually handle automated loops so devs don't get burned and now I'm locked up for a week unless I pay even more money and my account shows about **10 billion tokens** burnt. That's not fair, not to me, not to the economy, not to the other people who are running these challenges. Do you even know the electricity cost? Your services need to be more stable if you want to sell intelligence to humanity.
document fetched 2026-10-14 00:28 · canonical https://reddit.com/r/OpenAI/comments/1x4n1mk/the_openai_handshake_challenge_broke_my_agent