Claude Code talking about unexpected, different projects

27 points by WeaselsWin ↗ HN
Anyone else having this just now? Active Claude Code session spewing out responses about clearly someone else's prompts.

11 comments

[ 4.0 ms ] story [ 27.7 ms ] thread
(comment deleted)
it's no less likely to hallucinate than anything else, but you should take a screenshot
Context overflow so it’s hallucinating like a drunk sailor?
Is this just a result of aggressive caching? Like, "ya this code kinda seems like this other one I've computed, lemme just serve this past output"
Hey, WeaselsWin is my colleague and i'm the one who had this weird hallucination. I was in plan mode and i confirmed the plan and suddenly Claude Code started outputting text about completely different topics and hallucinating like this:

see that the database system is already quite well understood in the instructions. Let me provide a brief summary:

  Understanding the Mission System                                                                                                                                   
                                         
  The mission system consists of several interconnected components that work together to manage tasks:                                                               
                                                                                                                                                                     
  Key Components

  ┌─────────────────────────┬─────────────────────────────────────────────────────────────────────────────────────────┐
  │        Component        │                                       Description                                       │
  ├─────────────────────────┼─────────────────────────────────────────────────────────────────────────────────────────┤
  │ Database (Total)        │ A persistent storage layer that maintains the full record of all tasks and their states │
  ├─────────────────────────┼─────────────────────────────────────────────────────────────────────────────────────────┤
  │ Mission Active (active) │ A real-time view of currently running/pending missions                                  │
  ├─────────────────────────┼─────────────────────────────────────────────────────────────────────────────────────────┤
  │ Mission Service         │ Core service managing mission lifecycle (create, update, complete)                      │
  ├─────────────────────────┼─────────────────────────────────────────────────────────────────────────────────────────┤
  │ Mission Calendar        │ Scheduling system for time-based mission planning                                       │
  └─────────────────────────┴─────────────────────────────────────────────────────────────────────────────────────────┘

  How They Interact

  1. Mission Service creates and manages missions through their lifecycle
  2. Database stores all mission data persistently
  3. Active missions are tracked separately for quick access to current work
  4. Calendar provides scheduling capabilities for future missions

  Status Flow

  Created → Active → In Progress → Completed
                   ↘ Cancelled


  The syst...
Yes, this is very common. I have been testing claude code, opencode, and many other cli tools. One rrealization is that askinga model to "do it correctly" under the assumption that it would definitly follow that
It happened to me earlier today too. The response the machine gave me was completely different, as if it was a different codebase and a completely different task. It also had a few paths to some files and the username used for the `$HOME` directory was not mine. Local mode, not cloud or anything like that, and Auto mode.

Couple of "what the heck are you talking about"s, but because the attitude continued, I switched to Sonnet 4.5 and tried again, and this fixed the issue for me. I forgot to document it, and when I tried to export the original conversation for analysis it was gone.

But yeah, very confusing.