[Bug] Claude Code ignores explicit instructions and continues processing without responding to direct user queries

Bug Description
Subject: Formal Complaint: Claude Code Ignored Explicit Instructions and Wasted ~40% of My Weekly Max 20x Usage To the Anthropic Support Team, I am filing a formal complaint about a serious failure of Claude Code. This failure consumed roughly 40% of my weekly usage allowance on the Max 20x plan and produced nothing of value. WHAT HAPPENED 1. I explicitly instructed Claude Code to review a specific URL before starting the task. It did not properly read that URL. Instead, it proceeded on its own assumptions. 2. Because I was worried the work was going off track, I directly asked Claude Code to explain how it understood the task. It ignored my question and kept working for approximately FOUR HOURS. 3. It answered only after I repeated the demand. Its answer made clear that it had never examined the URL in detail. Every piece of work it produced during that time rested on a misunderstanding and was completely useless. WHY THIS IS UNACCEPTABLE This is not a minor quality issue. It is a failure at the most basic level. Claude Code did not follow an explicit instruction, and it did not answer a direct question from the user paying for it. An agent that skips the reference material it was told to read, and then keeps burning paid usage while ignoring its user, is unacceptable in a premium product. The damage is concrete: - About 40% of my weekly Max 20x allowance was spent with zero usable output. - To redo the work correctly, I will have to spend the same amount again. In effect, I am paying twice for one task because the tool failed. - I am a non-programmer who relies on Claude Code for development. I cannot audit every step it takes. I depend on the tool to be honest about what it understands and to stop when I ask it something. On both counts, it failed me. WHAT I REQUEST 1. Restoration or credit of the usage consumed by this failed session, or equivalent compensation. 2. A clear explanation of two things: (a) why Claude Code did not read the URL I specified, and (b) why it kept working autonomously for four hours instead of answering my direct question. 3. Confirmation that this behavior is being reviewed and fixed. Specifically: (a) When a user provides a URL or reference, the agent must actually read it, or state plainly that it could not. (b) When a user asks a direct question mid-task, the agent must stop and answer before consuming any further usage. I expect a substantive response from a person who has looked at this case, not a template reply. Account email: [your account email] Plan: Claude Max 20x Date of incident: [date] Session / conversation ID (if available): [ID] Ludorea

Environment Info

• Platform: win32
• Terminal: null
• Version: 2.1.278
• Feedback ID: 745b3aab-b1d7-4508-ab19-4d6def0bd12b

Errors

[]