[MODEL] Sonnet 4.6

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

Update some documents in my Strapi 5 CMS

What Claude Actually Did

Claude ignored my command to only interact with the CMS using a library I developed for that purpose and instead wrote directly against the API. My code was built knowing Strapi 5 bugs and quirkiness. Instead, claude's code, which he applied directly to production, made enormous damage to the document repository.

Expected Behavior

Claude writing production-bound code for third-party APIs (Strapi v5) with false confidence despite knowledge gaps, without flagging uncertainty or recommending local testing first. That's a concrete, actionable behavior to improve — not just "got something wrong. It gave me the impression that it knew that system, replied to a number of my questions with authority although it really knew nothing about it.

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

This confidence problem is omnipresent. It seems to completely overlook that the systems that it uses have been developed by humans, that you cannot blindly trust API documentation, or even that systems will react as expected. It's a very pervasive issue.

Claude Model

Sonnet

Relevant Conversation

Impact

Critical - Data loss or corrupted project

Claude Code Version

claude code 2.1.96

Platform

Anthropic API

Additional Context

My CMS has over 1000 documents. Repairing has taken 8 weeks and costed me over $400 in api tokens to research and fix the issues. 8 weeks later, my website is still not back on.