[MODEL]

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 modified files I didn't ask it to modify

What You Asked Claude to Do

basic coding tasks...
Since late march 2026, Claude Code has become unusable for any coding work.

What Claude Actually Did

agents drifting outside contracted slice boundaries, adding unrequested scope, ignoring explicit non-goals, falsely reporting progress...... the list goes on.

Expected Behavior

at minimum intern level of code execution. ATP, I would not even trust Opus 4.6 with a coffee order... I literally have one instance running only to automate the commit process.

Files Affected

everything Claude code is tasked on. fresh instance, informed instance, guard-railed instance does not matter how much tooling exits or if Marcel Proust provided the prompt.

Permission Mode

Accept Edits was OFF (manual approval required)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

type something into a Claude Code prompt then expect a logical contribution to a task.

Claude Model

Opus

Relevant Conversation

Claude said, I'll do X (simple task), Claude does crazy S*&t then apologizes, admits odd behavior then does it again. we are talking "car keys in the freezer" level of work product with basic tasks.

Impact

Critical - Data loss or corrupted project

Claude Code Version

Version: 2.1.87 Session name: /rename to add a name Session ID: 2eff7e37-8db6-42f1-a3ef-2d05e407808a cwd: /home/bfadmin/Documents/datum-eda Login method: Claude Max Account Model: Default Opus 4.6 with 1M context · Most capable for complex work MCP servers: 1 connected, 2 need auth · /mcp Setting sources: User settings, Project local settings Esc to cancel

Platform

Anthropic API

Additional Context

does not seem to matter the task complexity, code base, language etc...