[Bug] Claude Code requires excessive confirmation prompts for routine actions #103830

Activity

1. github-actions

   added

   bugSomething isn't workingSomething isn't working

   platform:windowsIssue specifically occurs on WindowsIssue specifically occurs on Windows

   area:permissions

   needs-infoIssue needs more information to be actionableIssue needs more information to be actionable

   on Oct 20, 2026
2. JimmySpivey commented on Oct 21, 2026

   

   JimmySpivey

   on Oct 21, 2026

   Author

   More actions

   For context, I just want it to pick up and automate some modding for en UE5 game. There are custom scripts, everything is setup I have done this flow, the choices I would make are very obvious it doesn't take much reasoning/expertise... yet opus 5.5 and claude in general acts sooooo confused about any OBVIOUS decision, even when forewarned and holding its hand upfront.
3. Kobel123 commented on Oct 21, 2026

   

   Kobel123

   on Oct 21, 2026

   More actions

   Two different things can cause this, and they have different fixes.

   If it's the "Do you want to proceed?" prompts: those are permission prompts, and you can pre-approve your own scripts so they stop asking. Easiest way: when a prompt shows up for one of your scripts, pick "Yes, and don't ask again". That saves a rule for the repo. Or type /permissions and add the rules yourself, or put them in C:\Users\<you>\.claude\settings.json so they apply to every project:

   {
     "permissions": {
       "allow": [
         "PowerShell(./scripts/*)",
         "Bash(python tools/*)"
       ]
     }
   }

   Change the patterns to match the command text you see in the prompt, otherwise they won't match. For file edits, the acceptEdits mode stops those prompts (Shift+Tab in the terminal, or the mode selector in the desktop app).

   If it's Claude asking you about decisions: that part isn't permissions, so the settings above won't help. Put the obvious choices in a CLAUDE.md file in the project folder, for example "run build.ps1 after every change, don't ask which mod folder to use, it's always Mods/MyMod". Claude reads that file at the start of every session.
4. JimmySpivey commented on Oct 23, 2026

   

   JimmySpivey

   on Oct 23, 2026

   Author

   More actions

   FYI: These are issues with the training of the LLM itself, not harness issues / lack of harness usage knowledge. The LLM should have "reasoned" / behaved such that it followed instructions. CLAUDE.md isn't the right approach for this one because it is a new project of some mods I just want to run it, not hold its hand. It acts like the modding setup is some sacred production live running app that it can't touch. Anyone using basic reasoning would infer this is a simple instruction mandate to reason through with a rather clear outcome, but the model training is over-tuned to be timid and acts confused when its reasoning tokens then sway it away from the original ask. It confused itself as well.
5. github-actions

   removed

   needs-infoIssue needs more information to be actionableIssue needs more information to be actionable

   on Oct 23, 2026