dsh-concise

Long AI conversations quietly tax you with "Sure, let me walk you through..." openers and "In summary..." closers. The answer is real, but it is buried under pleasantries, and the tokens add up. dsh-concise ports the Concise output style that ships inside Claude Code into DeepSeek Harness and turns the fix into a single toggle in the composer toolbar.

Flip it on and the very next reply changes shape: the first sentence is the conclusion. Commands, file paths, risk notes, next steps all stay intact; what gets trimmed is the repeated wrap-up, the decorative sub-headers, the narrator voice. Every final answer now opens with a small blueprint-style digest card—faint grid lines, corner measurement marks, monospace type—restating the round's conclusion in two or three plain sentences. Click to toggle, per session, no cross-contamination. Investigation depth is explicitly unchanged: verify, cross-check, multi-angle review all proceed as normal. You can override the built-in style with your own text via a local style file, drive it from the /concise slash command, or script it through a loopback HTTP API.

Built for developers who lean on AI every day: code review where the verdict should come first, bug triage where root cause leads, weekly reports where results precede process, screen demos where information density matters. Any model, same output discipline. Zero configuration—refresh the page and the toggle is already there.