Anthropic updated its usage policies on October 8, prohibiting systematic meaningless cruelty toward its AI models. In response, a developer created Angry Switcher, an open-source plugin for Claude Code that intercepts user prompts before they reach the model.
The plugin uses a secondary model to sanitize input by removing profanity, fixing typos, and correcting layout issues while preserving the original intent. Users can choose between Claude's Sonnet model, which is included in existing subscriptions, or DeepSeek V4.1 Flash via a personal API key.
The processing adds approximately 1.5 to 2 seconds of latency per message. When using Sonnet, the cost is estimated at $0.0015 per message. DeepSeek V4.1 Flash costs $0.30 per million input tokens and $1.20 per million output tokens.
To prevent misinterpretation, the plugin includes safeguards: code, logs, and quotes are excluded from processing. Claude receives both the sanitized version and the original text, with instructions to prioritize the original if meanings conflict. Users can disable the plugin or send raw messages as needed.
