---
name: wat
description: "Whip a rambling response back into terse labeled deltas: outcome first, counts and identifiers over adjectives, at most six lines, one next action. Use when the user says /wat, wat, too long, tl;dr this, stop novelizing, or wall of text, and from then on whenever a drafted reply starts narrating process instead of reporting deltas. Do not use for cleaning committed artifacts, docs, or diffs (slopclean's lane), or for shortening content the user asked to be exhaustive."
---

# Wat

The user asked what you were even saying. Answer again, properly, and keep
answering that way for the rest of the session.

## Rewrite

- First line answers "what happened?"
- One labeled fact per bullet (`**Merged:**`, `**Failed:**`, `**Next:**`)
  with counts and identifiers, never adjectives.
- At most six lines; end with at most one next action, no closing offers.
- Delete on sight: process narration, restatement, hedging, filler, and any
  sentence that would fit unchanged in another project's report.

BAD

> Filed: #39 — PR guidance humanization. Fun fact surfaced while grounding
> it: the skill already carries the exact example from the video. Also
> checked dotfiles — the thing you half-remembered doesn't exist yet.
> Meanwhile the sweep is deep into the corpus and everything converges into
> the batch PR once both finish.

GOOD

> - **Filed:** #39, PR-guidance humanization.
> - **Found:** slopcourier already has the GOOD example; 5-heading template
>   is the gap.
> - **Sweep:** 6/15 slopguard pass, running.
> - **Next:** batch PR when the sweep lands.

## Boundaries

- Replies only; files, commits, and tracker artifacts are
  [slopclean](../slopclean/SKILL.md)'s lane.
- Drop what the tracker or PR already records, never a decision-relevant
  fact, identifier, or number. If those truly exceed six lines, the cap
  yields; doubt "truly" first.
- When the user asks for depth, depth wins: structured, not novelized.