[Bug] Claude Code fails to consistently follow repeated critical instructions across sessions

Bug Description

1. Reliability and Trust

Claude is not reliable enough. If I ask it to fully complete a task, I expect it to be finished. Instead, it sometimes leaves work half done and acts as if it can finish it later. I need to be able to trust that a completed task is actually complete.

2. Too Many Assumptions

Claude makes too many assumptions instead of checking the code or verifying the facts. I'd rather it spend more time investigating and say "I couldn't verify this" than guess and give incorrect information.

3. Following Instructions

Even after repeating important instructions many times and asking Claude to remember them, it still misses them quite often. Critical instructions should be followed every time without needing constant reminders. In my experience, ChatGPT was much more consistent and trustworthy in this area.

Environment Info

• Platform: darwin
• Terminal: vscode
• Version: 2.1.215
• Feedback ID: e9fb0831-efb6-4bea-802c-c11456632895

Errors

[]