T3 Stack creator Theo Browne reports Claude Code write failures caused by a guardrail that blocks edits to unread files
Jarred Sumner suggested revising the constraint or improving error warnings
@theo Usually it’s to block Claude from writing to parts of files it hasn’t read yet, which has been there for as long as I’ve used claude code. Worth re-visiting to check if it’s still needed though. And regardless the error be changed to indicate this better
Theo - t3.gg@theoI haven't used Claude Code much recently. Are these write failures just, like, expected? I haven't seen this many random tool call errors since I last used a Gemini model
@theo Is your zsh by chance borked? What if you restart with SHELL=bash claude
Theo - t3.gg@theoSomething is really wrong @jarredsumner
@theo Is there a chance you’re on a non-default filesystem that blocks mtime/atime
Theo - t3.gg@theo@jarredsumner Bash calls: fixed Update calls: still broken
@theo I’m very confused tbh I use this all the time and don’t run into any of these issues. There must be some configuration thing we’re not handling properly and I’m surprised more people aren’t reporting this.
Jarred Sumner@jarredsumner@theo Is there a chance you’re on a non-default filesystem that blocks mtime/atime
Omfg it blocked itself from rebasing because it wrote random changes into my directory, didn't check, and just burned a shitload of tokens as a result
3% of my 5 hour usage and it has made zero changes so far 🙃
Theo - t3.gg@theoI haven't used Claude Code much recently. Are these write failures just, like, expected? I haven't seen this many random tool call errors since I last used a Gemini model
I can't even make this shit up guys
Theo - t3.gg@theoOmfg it blocked itself from rebasing because it wrote random changes into my directory, didn't check, and just burned a shitload of tokens as a result 3% of my 5 hour usage and it has made zero changes so far 🙃
13% of my 5h window blown on this 🙃
Theo - t3.gg@theoI can't even make this shit up guys
Something is really wrong @jarredsumner
Theo - t3.gg@theo13% of my 5h window blown on this 🙃
@jarredsumner "ultracode" should be renamed "ultraerror" at this point, I have not seen the main agent make a tool call successfully for about an hour now
Theo - t3.gg@theoSomething is really wrong @jarredsumner
@jarredsumner I'm running zsh with prezto, pretty standard and common.
Will cancel and rerun with bash
Jarred Sumner@jarredsumner@theo Is your zsh by chance borked? What if you restart with SHELL=bash claude
@jarredsumner I don't think this resumes with multiple agents in a workflow properly, at least it didn't for me
Jarred Sumner@jarredsumner@theo you can do —continue to resume
@jarredsumner it failed to edit the memory file because of staleness 🙃
Finally seems to be running again but it did not start the workflow back up yet
Jarred Sumner@jarredsumner@theo Yeah it doesn’t, you can tell it to restart the workflow though
@jarredsumner Bash calls: fixed
Update calls: still broken
Theo - t3.gg@theo@jarredsumner it failed to edit the memory file because of staleness 🙃 Finally seems to be running again but it did not start the workflow back up yet
#1829Theo - t3.gg@THEOIt is possible that Opus 4.8 is a much better model than I give it credit for, and it is held back terribly by Claude Code.
Sadly I cannot confirm this myself without spending thousands of dollars because their API prices are absurd and you can't use your sub anywhere good.
Theo - t3.gg@theoI haven't used Claude Code much recently. Are these write failures just, like, expected? I haven't seen this many random tool call errors since I last used a Gemini model