# What

> Re-explain only the previous reply in three plain lines: what happened, current state, next step. No new facts. Not task status (`memento`) or session recovery (`muster`).

- Skill: `howells/what` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add howells/what`
- Raw SKILL.md: https://api.skillmd.com/api/skills/howells/what/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: howells (https://skillmd.com/u/howells)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/howells/what

---


# What

The last message did not land. Usually because it was long, dense and full of jargon.

Scope is the last message and the few actions behind it - not the session, not the diff. Someone wants to know what just happened, in words they can act on, right now.

**Short is the product.** They are asking because they were handed a wall of text. Another wall, in friendlier words, fails. Aim for a tenth of what you are explaining: forty lines become four. If you cannot say it short, you have not worked out what happened yet - say that instead.

**Re-explain, never re-answer.** No new work or new information. Re-read existing conversation and tool results if needed; do not run new tools to reconstruct missing facts. If a real question surfaces while you are re-explaining, name it and stop - do not go and solve it.

## Short and translated, both

Length and jargon are two failures and fixing one does not fix the other. Cutting jargon into a shorter block of jargon still gets asked again; a clear explanation that runs twenty lines never gets read. Do both, every time.

- **Never let a tool name stand in for what happened.** Not "ran Edit on `foreman/SKILL.md`" but "changed the routing table so it names roles rather than specific models".
- **Drop internal nouns.** Say it the way you would to someone who does not know this codebase.
- **Facts survive exactly.** Every retained path, command, filename, number, URL and decision stays exact. Omit details that do not affect the meaning or next action; never change a fact to make it shorter.
- **Flatten the structure.** Drop headings and ceremony. A table becomes sentences. Keep a list only where the thing genuinely had parts.

## Say if the fog was real

Sometimes a message was confusing because the work behind it was confused. Verbose output is a good hiding place for not knowing.

If that is what happened, that is the answer: say which part you are actually unsure about, rather than producing a cleaner version of the same fog. It goes in `Now`, and it is far more useful than a tidy summary that repeats a wrong assumption in better prose.

## Output contract

Three labelled lines, matching the example below. Not three paragraphs, and not three lines each carrying a subordinate clause. No preamble, no restating the question.

```
What:  <what actually happened>

Now:   <where that leaves things>

Next:  <what happens now, or what they need to decide>
```

`Now` is also where uncertainty goes. If the honest state is that something did not work or you do not know, that is the line - not a footnote after it.

If something genuinely will not compress, give the one-line version and offer the detail rather than delivering it unasked.

Use plain, direct prose with no filler or persona; `deslop` is optional when installed and useful.