dsh-tacit: Learning What Your Prompts Leave Unsaid

`dsh-tacit` is a workflow plugin for DeepSeek Harness designed to address the issue of agents stalling due to unclear prompts. It observes user corrections and interaction history, automatically infers unspoken preferences, and converts them into system instructions injected into the conversation to reduce repeated corrections. The plugin supports zero-click learning, instruction probation management, draft rewriting, and visualization of interaction data. Regarding privacy, it uses local storage and model-call routing, does not directly access API keys, is licensed under MIT, and has a very low learning cost per instruction.