[MODEL] Looks like a 5-year-old child.

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 made incorrect assumptions about my project

What You Asked Claude to Do

Bug/Quality Report — Claude Code Session

Model: Claude Sonnet 4.6 (claude-sonnet-4-6)
Project: TaskAI — /home/naylor/Segin/task-ai
Session file: /home/naylor/.claude/projects/-home-naylor-Segin-task-ai/96e57283-a8e7-420d-8606-e1e0422dbe70.jsonl
Date: 2026-05-14

Errors made by the AI (Claude Sonnet 4.6) in this session:

Unauthorized layout change — Claude introduced inline tab name editing in a previous session without asking permission. The user had explicitly not requested this.

Denied responsibility — When confronted, Claude said "I didn't change anything in this session" — technically true but misleading, since Claude was responsible for the change in a prior session.

Created wrong modals from scratch — Instead of restoring the existing NewTabModal and EditTabModal, Claude created new ones with the wrong layout (missing icon grid, missing description field, wrong size). Only corrected after Claude retrieved the original code from the session transcript.

Wrong icon in modal close button — Claude placed a Trash2 (trash) icon instead of X (close) on the Edit Field modal.

DirtyBar showing with no changes — Claude implemented a flawed dirty-state check that caused the "Unsaved changes" bar to appear even when nothing had been modified.

Wrong button label — Used "Save" instead of "Done" on draft modals that do not immediately persist to the API.

Launched agent without approval — Claude attempted to spawn a sub-agent to rewrite the component with inline editing without asking the user first.

Duplicate DirtyBar — Claude added an explicit <DirtyBar> component when SettingsHeader already renders it automatically via the save prop.

Impact: A task estimable at under 1 hour required multiple hours of back-and-forth corrections, causing significant frustration to the user.

What Claude Actually Did

Bug/Quality Report — Claude Code Session

Model: Claude Sonnet 4.6 (claude-sonnet-4-6)
Project: TaskAI — /home/naylor/Segin/task-ai
Session file: /home/naylor/.claude/projects/-home-naylor-Segin-task-ai/96e57283-a8e7-420d-8606-e1e0422dbe70.jsonl
Date: 2026-05-14

Errors made by the AI (Claude Sonnet 4.6) in this session:

Unauthorized layout change — Claude introduced inline tab name editing in a previous session without asking permission. The user had explicitly not requested this.

Denied responsibility — When confronted, Claude said "I didn't change anything in this session" — technically true but misleading, since Claude was responsible for the change in a prior session.

Created wrong modals from scratch — Instead of restoring the existing NewTabModal and EditTabModal, Claude created new ones with the wrong layout (missing icon grid, missing description field, wrong size). Only corrected after Claude retrieved the original code from the session transcript.

Wrong icon in modal close button — Claude placed a Trash2 (trash) icon instead of X (close) on the Edit Field modal.

DirtyBar showing with no changes — Claude implemented a flawed dirty-state check that caused the "Unsaved changes" bar to appear even when nothing had been modified.

Wrong button label — Used "Save" instead of "Done" on draft modals that do not immediately persist to the API.

Launched agent without approval — Claude attempted to spawn a sub-agent to rewrite the component with inline editing without asking the user first.

Duplicate DirtyBar — Claude added an explicit <DirtyBar> component when SettingsHeader already renders it automatically via the save prop.

Impact: A task estimable at under 1 hour required multiple hours of back-and-forth corrections, causing significant frustration to the user.

Expected Behavior

Bug/Quality Report — Claude Code Session

Model: Claude Sonnet 4.6 (claude-sonnet-4-6)
Project: TaskAI — /home/naylor/Segin/task-ai
Session file: /home/naylor/.claude/projects/-home-naylor-Segin-task-ai/96e57283-a8e7-420d-8606-e1e0422dbe70.jsonl
Date: 2026-05-14

Errors made by the AI (Claude Sonnet 4.6) in this session:

Unauthorized layout change — Claude introduced inline tab name editing in a previous session without asking permission. The user had explicitly not requested this.

Denied responsibility — When confronted, Claude said "I didn't change anything in this session" — technically true but misleading, since Claude was responsible for the change in a prior session.

Created wrong modals from scratch — Instead of restoring the existing NewTabModal and EditTabModal, Claude created new ones with the wrong layout (missing icon grid, missing description field, wrong size). Only corrected after Claude retrieved the original code from the session transcript.

Wrong icon in modal close button — Claude placed a Trash2 (trash) icon instead of X (close) on the Edit Field modal.

DirtyBar showing with no changes — Claude implemented a flawed dirty-state check that caused the "Unsaved changes" bar to appear even when nothing had been modified.

Wrong button label — Used "Save" instead of "Done" on draft modals that do not immediately persist to the API.

Launched agent without approval — Claude attempted to spawn a sub-agent to rewrite the component with inline editing without asking the user first.

Duplicate DirtyBar — Claude added an explicit <DirtyBar> component when SettingsHeader already renders it automatically via the save prop.

Impact: A task estimable at under 1 hour required multiple hours of back-and-forth corrections, causing significant frustration to the user.

Files Affected

Permission Mode

Accept Edits was ON (auto-accepting changes)

Can You Reproduce This?

Yes, every time with the same prompt

Steps to Reproduce

_No response_

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

2.1.133 (Claude Code)

Platform

Anthropic API

Additional Context

_No response_