ai-input-enhancer

Writing a sloppy prompt to an LLM usually drags the quality of the answer down, but manually polishing every sentence before hitting send is slow and breaks your flow. ai-input-enhancer adds a Prompt Optimize button right inside the DeepSeek Harness input box: click it, the current draft is rewritten by the model you have configured, the result lands back in the box, and one click reverts you to the original text. Everything stays in the conversation—no separate editor, no copy-paste round trip.

Three clearly bounded optimization modes give you control without guesswork. Fidelity improves clarity without inventing requirements, facts, or scope. Dev-oriented reformats the technical details you already provided into implementation-ready language, without picking a stack or architecture on your behalf. Spec Expansion may fill in necessary defaults, but every assumption is explicitly listed under a Default assumptions heading for your review. On the context side, the plugin supports an optional short-term memory of up to three applied optimizations within the session, and optional Markdown documents from the current workspace root. Both sources are subject to a configurable character budget and degrade gracefully if the underlying service is unavailable. You can also configure the primary model, fallback model, timeout, and max output tokens through a segmented advanced-settings panel.

This plugin is best suited for developers and product folks who rely on DSH for technical Q&A, design discussions, or spec writing and want the discipline of optimizing before sending while keeping tight control over what the model may add. Privacy-wise, the plugin never persists full drafts, conversation history, or document contents, and it does not store API keys. All extra context sources are off by default—you turn them on one at a time.