Got nuked for hitting $5k in a single day of using claude code
I was really liking the workflow feature of claude code wherein 1000s of subagents could be reliabily spawned and managed for a complex long running task.
I used the workflow in ultracode mode to find security vulnerabilities similar to a Mythos scan a Rust and C repo function by function rather than looking at once by creating a threat model first and then having independent subagents running falsification tests, reviews, PoC reproduction for possible vulnerabilities using Opus, Sonnet and Haiku. So that I could compare results and know unique issues found by all of them.
Long story short, after I approved the plan, and then the session spawned close to 40k subagents to peform the work for the repo and sent about 310k requests and I landed up with a usage bill of $5k. My company straight-up suspended my account, now I can't use claude, I am lucky they are not futher escalating the issue about token usage😅
Would appreciate if others can their story as well, so I don't feel alone.