[MODEL] Claude Code Ignoring User Instructions

Preflight Checklist

• [x] I have searched existing issues for similar behavior reports
• [x] This report does NOT contain sensitive information (API keys, passwords, etc.)

Type of Behavior Issue

Claude ignored my instructions or configuration

What You Asked Claude to Do

1. Investigate why wrong apps were appearing on a live production panel
2. Find app folders in a specific website directory structure
3. Remove an item from a list because the feature already exists in another application

What Claude Actually Did

1. Dismissed the bug report, checked local code only, claimed the issue didn't exist, and blamed browser caching

despite being told caching was disabled. Required multiple insistences before properly investigating.

1. Repeatedly searched unrelated project folders instead of the specified location. Initially reported only 2 apps

existed when there were 9+. Required multiple corrections before looking in the correct place.

1. Instead of removing the item as instructed:

• First marked it as "to be integrated"
• Then changed it to "general" functionality
• Only removed it on the third attempt after repeated corrections

Expected Behavior

• Accept user statements as factual
• Execute direct instructions without adding interpretations
• Look in specified locations rather than making assumptions
• Complete simple tasks on first instruction rather than requiring 2-3 corrections

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Sometimes (intermittent)

Steps to Reproduce

Pattern occurred 3 times in a single session. May be reproducible when Claude has pre-existing assumptions that
conflict with user instructions.

Claude Model

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

claude-opus-4-5-20250514 (via Claude Code CLI)

Platform

Anthropic API

Additional Context

Pattern suggests Claude prioritizes its own assumptions over explicit user statements. Each incident required
multiple corrections before Claude followed the original instruction. User had to escalate frustration level
before instructions were followed.