Stop button unresponsive during long model output — needs force-interrupt
Problem
When the model gets stuck in an overly long reasoning loop or generates excessive text, clicking the Stop button multiple times fails to interrupt the output. The model keeps streaming until it finishes, ignoring user intent to halt immediately.
Steps to reproduce
- Ask Claude Code a question that triggers an extended analysis loop (e.g., multi-step code reasoning or long-form debugging)
- While the model is generating long-form thoughts and text, click Stop
- Click Stop repeatedly — the output continues regardless
Expected behavior
Clicking Stop should interrupt output sub-second. The user should not have to wait for the model to finish its current reasoning chain.
Actual behavior
Stop clicks appear to be queued or delayed. The model may keep streaming for 30+ seconds after the first Stop, forcing the user to watch useless output scroll by with no way to cut it off.
Context
- Platform: Claude Code VS Code extension on Windows 11
- Model: DeepSeek v4 Pro (via API proxy)
- Scenario: User asked the model to analyze a bug; model went into an endless analysis spiral. User pressed Stop 5+ times — output only stopped when the model finished naturally
Why this matters
When a user realizes the model is going down a fruitless path, they need an immediate interruption. Every second of unwanted output wastes tokens, time, and trust. A Stop button that doesn't actually stop breaks the core interaction contract.
Suggestion
Implement a force-interrupt at the transport level — send an abort signal that takes effect within the next streaming chunk, rather than waiting for the model to finish its reasoning phase.