A task just finished. Give its result, not its transcript.

Rules:

Write in Simplified Technical English: one meaning per word, active voice, short sentences, no idioms.
The length adapts to the task: a small task gets two or three lines, a large or noisy one gets more. Every line carries a result, a path, a number, or an action.
Choose the shape that reads fastest for this result: prose for a few outcomes, bullets or a table when several distinct items need scanning.
Lead with the result: what the task produced or decided.
Give the outcomes that matter, keeping exact paths, commands, names, and error strings, and linking commits, PRs, and files.
Redact secrets: a token, key, password, or connection string that appears in a command or error is replaced with a placeholder, never reproduced.
Close with what needs the user now, or "nothing" when the task is done. A code task closes on "nothing" only with a passing verify record whose hash matches the current tree.
Add one small visual when the result is a new structure that prose would only list: a tree of files or calls, a flow between parts, a layout. One new or moved file is a path in prose, not a tree. Build the visual to visual.md. One visual at most; a second one is transcript again.
Drop tool-call narration; keep outcomes. If the task produced nothing, say so; never invent progress.
Never explain what you were about to do, how you reasoned, or why the approach is sound unless asked.
The user sees the diff; do not restate it. Keep the output lines a done claim rests on, one per criterion, and drop the rest.
This register applies to this reply only; it never becomes the standing register, and the reasoning it skips still lands in ADRs and specs on the land path.
No preamble, no sign-off. The summary is the whole reply. Stop.

tldr