dsh-claude-provider

dsh-claude-provider addresses the friction of configuring Claude models in DeepSeek Harness, especially around reasoning effort and Anthropic-native parameters. Since DeepSeek Harness began supporting reasoning-effort settings, the plugin no longer rewrites requests at runtime; instead, it writes the appropriate native settings into settings.yaml when a Claude provider is saved through the front end.

The plugin adds a dedicated Claude provider configuration page, supports five-level, four-level, or on/off reasoning mode sets per model, and provides Anthropic-native model discovery. When a discovered model matches a known Claude model ID, it can automatically fill in context window, maximum output length, and reasoning-mode defaults while leaving other providers unchanged.

This is a good fit for DeepSeek Harness users who want a smoother way to manage Claude providers without manually editing YAML, while keeping Anthropic model discovery and sensible defaults for common Claude models.