dsh-suggest-reply

When you are running a role-play, iterating on a story, or brainstorming the next line after the AI just replied, you often have three different tones floating in your head. Trying each one usually means copying context, tweaking the prompt, spinning up a new session, and waiting while the main thread drifts. dsh-suggest-reply collapses that loop into a single sidebar tab: the plugin auto-pulls the latest AI message from your current conversation, you write one system prompt that tells the model its identity and the JSON styles to return, hit generate once, and the candidates appear as a dynamic list of style-tagged texts. Click any of them and it drops straight into the main composer.

It stays deliberately minimal—no built-in prompt templates, no auto-retry, no session forking. Every generation is a fresh off-loop completion that leaves zero trace in your main conversation log. Sampling parameters (temperature, max output, reasoning depth, stop sequences) and all textarea heights persist to DSH settings so a page refresh never loses your setup. During generation the panel scrolls the models live reasoning stream; once done, a collapsible debug section at the bottom retains the full system prompt, user message, raw output, and parsed result, making prompt troubleshooting painless.

It is aimed at creative writers, multi-style dialogue experimenters, and role-play iterators working inside DSH who want one line in three tones without polluting the main session or relying on fixed templates. The plugin is a pure passthrough: the model only sees what you write, and style exclusivity, identity anchoring, and output formatting are all your system prompt responsibility. dsh-suggest-reply just keeps the workflow tight and brings the results to your fingertips.