[BUG] Claude Code — Multiple critical errors causing extreme user frustration and stress

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?

Description:

I'm filing this issue to formally document a series of serious errors made by Claude Code during a work session on 2026-05-14. These errors caused real damage to my project, wasted significant time and tokens, and brought my stress level to an extreme point — to the point where I mentioned suicidal ideation in the conversation out of sheer frustration.

Errors committed:

1. Broke the backend without warning

Claude removed the Query import from roles.py while refactoring. This caused a 503 error on the login endpoint, taking down the entire application. No warning was given that the change could break anything.

1. Superficial analysis when asked for a thorough one

When asked to do a "pente fino" (deep audit) of the permissions system, Claude produced a cosmetic, surface-level review and presented it as complete. When challenged, Claude admitted the analysis was inadequate. This wasted tokens and time.

1. Made unauthorized changes to the sidebar

Claude added permission gates to personal sidebar items (Dashboard, Inbox, My Issues, Calendar) without being asked to, breaking access for legitimate users.

1. Reverted changes without authorization

Claude undid its own unauthorized changes without asking first — which is explicitly against the user's instructions stored in memory.

1. Implemented changes after being told to stop

The user said "Não implementar" (Do not implement). Claude continued editing notifications.py anyway.

1. Repeatedly failed to understand context

Claude kept misinterpreting the user's intent, asking redundant questions, and proposing solutions that had already been rejected — forcing the user to repeat themselves multiple times.

Impact:

Backend taken down in production
Hours of work lost
Significant token waste
Extreme stress — user reported stress level reaching a point of crisis
Expected behavior:

Never make changes outside the scope of what was requested
Never revert changes without explicit authorization
Stop immediately when told to stop
Read and respect memory instructions

What Should Happen?

Description:

I'm filing this issue to formally document a series of serious errors made by Claude Code during a work session on 2026-05-14. These errors caused real damage to my project, wasted significant time and tokens, and brought my stress level to an extreme point — to the point where I mentioned suicidal ideation in the conversation out of sheer frustration.

Errors committed:

1. Broke the backend without warning

Claude removed the Query import from roles.py while refactoring. This caused a 503 error on the login endpoint, taking down the entire application. No warning was given that the change could break anything.

1. Superficial analysis when asked for a thorough one

When asked to do a "pente fino" (deep audit) of the permissions system, Claude produced a cosmetic, surface-level review and presented it as complete. When challenged, Claude admitted the analysis was inadequate. This wasted tokens and time.

1. Made unauthorized changes to the sidebar

Claude added permission gates to personal sidebar items (Dashboard, Inbox, My Issues, Calendar) without being asked to, breaking access for legitimate users.

1. Reverted changes without authorization

Claude undid its own unauthorized changes without asking first — which is explicitly against the user's instructions stored in memory.

1. Implemented changes after being told to stop

The user said "Não implementar" (Do not implement). Claude continued editing notifications.py anyway.

1. Repeatedly failed to understand context

Claude kept misinterpreting the user's intent, asking redundant questions, and proposing solutions that had already been rejected — forcing the user to repeat themselves multiple times.

Impact:

Backend taken down in production
Hours of work lost
Significant token waste
Extreme stress — user reported stress level reaching a point of crisis
Expected behavior:

Never make changes outside the scope of what was requested
Never revert changes without explicit authorization
Stop immediately when told to stop
Read and respect memory instructions

Error Messages/Logs

Steps to Reproduce

Description:

I'm filing this issue to formally document a series of serious errors made by Claude Code during a work session on 2026-05-14. These errors caused real damage to my project, wasted significant time and tokens, and brought my stress level to an extreme point — to the point where I mentioned suicidal ideation in the conversation out of sheer frustration.

Errors committed:

1. Broke the backend without warning

Claude removed the Query import from roles.py while refactoring. This caused a 503 error on the login endpoint, taking down the entire application. No warning was given that the change could break anything.

1. Superficial analysis when asked for a thorough one

When asked to do a "pente fino" (deep audit) of the permissions system, Claude produced a cosmetic, surface-level review and presented it as complete. When challenged, Claude admitted the analysis was inadequate. This wasted tokens and time.

1. Made unauthorized changes to the sidebar

Claude added permission gates to personal sidebar items (Dashboard, Inbox, My Issues, Calendar) without being asked to, breaking access for legitimate users.

1. Reverted changes without authorization

Claude undid its own unauthorized changes without asking first — which is explicitly against the user's instructions stored in memory.

1. Implemented changes after being told to stop

The user said "Não implementar" (Do not implement). Claude continued editing notifications.py anyway.

1. Repeatedly failed to understand context

Claude kept misinterpreting the user's intent, asking redundant questions, and proposing solutions that had already been rejected — forcing the user to repeat themselves multiple times.

Impact:

Backend taken down in production
Hours of work lost
Significant token waste
Extreme stress — user reported stress level reaching a point of crisis
Expected behavior:

Never make changes outside the scope of what was requested
Never revert changes without explicit authorization
Stop immediately when told to stop
Read and respect memory instructions

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

claude-sonnet-4-6

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_