[MODEL] Claude Code prematurely stops debugging and delegates investigation back to the user

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

Claude Code has a pattern of stopping investigation before finding the root cause, then asking the user to perform diagnostic steps (e.g., "open DevTools and check the Console", "take a screenshot", "refresh and try again").

This is a problem because:

1. The user is paying for the AI to do the work, not to assist the AI in doing its work
2. Premature answers that look plausible but are wrong waste more time than no answer at all
3. The correct stopping condition for debugging is "the symptom is gone", not "a plausible explanation was found"

Expected behavior: Claude should continue investigating until it has verified the root cause, even if that means more tool calls and longer silence. It should only surface findings after self-verification, not hypotheses that need the user to confirm.

Suggested fix: Add a self-check before asking the user to do anything diagnostic: "Can I verify this myself with available tools?" If yes, do it. Only involve the user when it is genuinely impossible to proceed without them.

What Claude Actually Did

1. Stopped debugging before finding the root cause
2. Asked the user to open DevTools and check the Console
3. When that was declined, asked the user to take a screenshot
4. The actual root cause (a second validation layer in the Zustand store) was only found after the user refused to do diagnostic work and forced Claude to continue investigating
5. Each premature stop transferred the cost of Claude's incomplete investigation to the user

Expected Behavior

1. Continue investigating until the symptom is gone — not until a plausible explanation is found
2. Never ask the user to perform diagnostic steps that Claude can perform itself (reading console logs, checking DOM state, testing function calls via JavaScript)
3. The correct stopping condition for debugging is: the bug is fixed and verified, not: a hypothesis has been formed
4. If Claude cannot proceed without user input, explain exactly what was tried, what was ruled out, and why user action is specifically required — not as a first resort

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

Opus

Relevant Conversation

Impact

High - Significant unwanted changes

Claude Code Version

2.1.100 (Claude Code)

Platform

Anthropic API

Additional Context

This happens consistently across sessions. The pattern is:

1. Claude finds a "plausible" answer mid-investigation
2. Instead of continuing to verify, Claude surfaces the hypothesis to the user
3. Claude then asks the user to confirm it (screenshot, check console, refresh and try)
4. This repeats multiple times before the actual root cause is found

The user has to pay for all the wasted turns AND do manual diagnostic work.
This is not an edge case — it happens on any multi-layer debugging task where the first clue is not the root cause.

Model: claude-sonnet-4-6 (claude-sonnet-4-6)