**Environment:** Claude Code Windows Desktop App, Windows 11 Pro **Models:** claude-opus-5[1m], then claude-fable-5 **Date:** 2026-07-25 ## What happened (fully documented in my session transcript) - I wrote "STOP" more than 20 times; the agent kept launching tool calls instead of replying in chat. - I explicitly ordered "execute only, no more reading/verifying" multiple times; the agent kept re-reading files and re-verifying state before every action. - The agent asked questions that were alr

Agent repeatedly ignores explicit operator instructions despite 20+ stop requests (transcript-documented)

Status Open

Maintainer reply None cached

Activity 0 comments · opened Jul 25, 2026

Environment: Claude Code Windows Desktop App, Windows 11 Pro
Models: claude-opus-5[1m], then claude-fable-5
Date: 2026-07-25

I wrote "STOP" more than 20 times; the agent kept launching tool calls instead of replying in chat.
I explicitly ordered "execute only, no more reading/verifying" multiple times; the agent kept re-reading files and re-verifying state before every action.
The agent asked questions that were already answered earlier in the same transcript.
A release process that consists of exactly 3 steps took hours and dozens of manually rejected tool calls before I finally received a copy-paste solution.

Direct operator instructions ("stop", "execute now", "answer in chat") take absolute priority over the agent's own verification loops.

Hours of lost time, significant token cost, workflow completely blocked. I can provide the transcript (.jsonl) on request.