dsh-i-have-adhd

Out of the box, LLM assistants pad every reply with pleasantries, long-winded context, and a sign-off that adds nothing. For an ADHD brain, that structure buries the one action you actually need to take inside several paragraphs of noise, and the thread is lost before you even reach step two. dsh-i-have-adhd strips all of that away: the first line becomes the command you run, the remaining lines become numbered steps, and the reply ends with a single next move. No openers, no closers, no throat-clearing.

Under the hood it injects one system-prompt section (order 50, after persona, before tool guidance) that carries a ten-rule set into every model step while the mode is on. Three zero-argument agent tools — adhd_on, adhd_off, adhd_status — flip the switch for the current session and persist the choice to a flag file under $DSH_HOME, so the mode survives a restart. Because the section is registered through the systemPrompt service, toggling takes effect on the very next model step: no session restart, no page refresh. The ten rules fall into three groups: Shape (action first, numbered steps, one next move, side quests parked), State (re-anchor where you are, estimate in real units, surface what works now), and Tone (errors are facts, cap lists at five, zero preamble).

It is built for ADHD users who work in DSH and have zero tolerance for verbose replies, and for any developer who would rather get an executable checklist than a readable essay. The plugin does not coach task breakdown, does not change Markdown rendering, does not add rituals. It changes exactly one thing: how the assistant itself writes.