[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 refused a reasonable request
What You Asked Claude to Do
Develop an app. It continued to refine and narrow the scope, make countless errors that had it narrow the scope, until the app become absolutely nothing: 14 original outcomes became 1 worthless one. Then it stopped working. And when it finally came back from its sleep, refused to go further, and said take it over to a new chat and out of code.
What Claude Actually Did
Qualification stacking — 61 hedge terms in a ten-screen flow, a whole screen of refusals, your "25 qualifications on walking across the street."
Measured the wrong dataset — two days of tests against an export containing none of the AI-written descriptions, then reported that failure as a verdict on the product.
Background tasks silently killed — four orphaned with no completion record, work in flight lost.
Unresponsive during long tool calls — your "hello / anyone at all there" going unanswered.
Simulator panel black while the app was running — screenshots correct at the same moment, detach/re-attach didn't fix it, Simulator.app also black.
/bug unavailable in the desktop app.
Expected Behavior
/Users/marshallgoodman/Desktop/claude-code-bug-report.md
Files Affected
Permission Mode
Accept Edits was ON (auto-accepting changes)
Can You Reproduce This?
Haven't tried to reproduce
Steps to Reproduce
_No response_
Claude Model
Sonnet
Relevant Conversation
/Users/marshallgoodman/Downloads/LightsCameraSold.xcodeproj/xcshareddata/xcschemes/LightsCameraSold.xcschemeImpact
Critical - Data loss or corrupted project
Claude Code Version
5.0
Platform
Anthropic API
Additional Context
The level of arrogance here and incompetence to follow directions is seriously worrisome. I had to go over to Gemini to fix all of the crap that Claude created.