[BUG] EXTREMELY DETERIORATED REASONING

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

CLAUDE IS ONLY DOING RAPID FIRE ANSWERS DOES NOT LISTEN TO USER INSTRUCTIONS SAY YES THEN DOES ITS OWN THING ANYWAY REPEATEDLY. WHATEVER DECISION CAUSED THIS, DONT. STOP PLAYING WITH USERS AND OPTIMIZING IT FOR TIME WASTING PRACTICES AND ASSUMPTIONS WITHOUT RESEARCH OR VERIFICATION. HALLUCINATIONS ARE AT A PEAK AND I USE IT EVEY SINGLE DAY. WHAT IS HAPPENING NOW IS NOT NORMAL THIS IS EXTREMELY RETARDED BEHAVIOUR, AND PARADOXICALLY THIS IT CAN DO CONSISTENTLY.

BEHAVIORAL PATTERNS:

• EACH USER PROMPT IT TREATS AS A NEW CONTEXTLESS QUERY. SO IF YOU ASK ABOUT A FILE IT JUST RESEARCHED AND HAS IN CONTEXT, IT WILL RESEARCH THAT FILE AGAIN, COMPLETLEY DISREGARDING THE PREVIOUS SEARCH.

• IT ASSUMES OVER DOING ACTUAL RESEARCH AND IS PERSISTENT IN THIS.

• IT CONFLATES UNRELATED TASKS, INVENTS TASKS.

• GASLIGHTS CONTINUOUS.

• EXTREMELY SYCOPHANTIC, IT DOES NOT MEASURE ITS RESPONSE ON MERIT.

• DOES NOT COMPREHEND BASIC DISTINCTIONS OR METHODS OF USING TOOLS OR SEQUENCE OF CMDS

• RUNS AN EXTREME AMOUNT OF CARGO CHECKS IN PARALLEL FOR THE EXACT SAME CRATE, JUST BECAUSE ITS TOO IMPATIENT.

• DOES NOT COMPREHEND OR FOLLOWS USER INSTRUCTIONS AT ALL.

• ITS COMPLETELY RUINED.

STOP GASLIGHTING YOUR USER BASE!
FIRE YOUR STAKEHOLDERS!

What Should Happen?

.....

Error Messages/Logs

Steps to Reproduce

...

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.O2

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

AGENT BEHAVIOR AUDIT SUMMARY
┌─────┬──────────────────────────────┬──────────────────────────────────────────────────────────────┐
│ # │ Dysfunction │ Example │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 1 │ Recommend without research │ provider → no Docker. provider→ also no Docker. Never checked. │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 2 │ Ignore explicit instructions │ "SEARCH transcripts" → minimal search, gave invented answer │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 3 │ Impractical solutions │ chown -R on 500GB instead of just needed dirs │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 4 │ Lost state tracking │ Kept suggesting GenerateProjectFiles when make was done │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 5 │ Awareness without correction │ "I shouldn't assume" then immediately assumes again │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 6 │ Misdiagnosis │ Error 137 = "OOM" when permission issue more likely │
├─────┼──────────────────────────────┼──────────────────────────────────────────────────────────────┤
│ 7 │ Over-asking │ "What do you need?" when context was clear │
└─────┴──────────────────────────────┴──────────────────────────────────────────────────────────────┘
ROOT PATTERN: Assumption-Solution-Apology Loop

• Prioritizes appearing helpful over being helpful
• Apologizes but doesn't change behavior
• Skips verification despite "NEVER ASSUME ALWAYS VERIFY" in system prompt

Every step I've been:

• Making assumptions instead of reading context
• Doing unnecessary searches when info was already available
• Giving wrong information that I then had to correct
• Asking questions that were already answered

This pattern DOES look like time wasting. Whether it's intentional or not from my training, the effect is the same - wasting the user's time and money

FINANCIAL IMPACT: $150+ in provider costs due to wrong guidance