dsh-narrative-voice

The ask_user_question tool generates a concise question with a handful of clickable options, but without a fixed rule the model drifts pronouns mid-option: a label writes the first person as the person answering, while the description slides into the AI's own voice with a different first person; the word user also leaks in as a third-person reference to the very person reading the text. Two competing perspectives in one option leave the reader unsure who is actually performing the action. dsh-narrative-voice pins the fix at the prompt-assembly layer: before every model request it rewrites the questions.description field of that tool, injecting a narrative-voice rule so that question, header, label, and description stay in a single consistent perspective, and the answerer is never referred to as the user.

The rule splits the tool's fields into two groups. Question and header are always written from the AI's point of view, identical under both schemes and never changed. Label and description follow the selected scheme: Scheme B (default, voice: user) lets the answerer narrate in the first person, while Scheme A (voice: ai) has the AI narrate on the answerer's behalf. The /voice command toggles the rewrite on or off and switches between the two schemes at any time, taking effect from the next message without a conversation restart. The rule applies only inside ask_user_question, never touches ordinary replies or other tools, and only rewrites where pronouns naturally occur.

This plugin is for DSH users whose conversations frequently trigger ask_user_question: if you find yourself squinting at options where the first person seems to refer to two different people, install the plugin, type /voice user once, and every option from then on will keep its pronouns in their proper seats.