Question 1

What does the wait-what skill actually do?

Accepted Answer

You type /wait-what, and the agent re-pitches its last message: it adds the context you were missing, drops the jargon, writes in ASD-STE100 Simplified Technical English, and uses the project's vocabulary from CONTEXT.md. The body is three lines long on purpose. The whole skill is a single leading word — 'wait' — plus instructions about how to re-explain.

Question 2

Why is the skill only three lines long?

Accepted Answer

The design doc says it plainly: skills that fight verbosity fail by growing. A 400-line concision skill still leaves the model verbose, because the model reads the volume, not the plea. Three lines carry one precise instruction and nothing else. The length is the mechanism, not a placeholder.

Question 3

Why does it say 'wait' instead of 'be concise'?

Accepted Answer

Because 'be concise' names the agent's output, and the model obeys by clipping words — which loses you further. 'Wait' names your state: comprehension failed here. An agent that hears 'be brief' writes telegrams. An agent that hears 'wait, you lost me' backs up and explains. The skill even names the counter-examples: /tldr, /no-fluff, and /talk-normal all over-correct into a caveman register that is shorter and no clearer.

Question 4

What is ASD-STE100 Simplified Technical English?

Accepted Answer

It's the aerospace maintenance writing standard — a controlled vocabulary and grammar rules designed so non-native speakers read maintenance manuals without ambiguity. The skill borrows it as the register for re-pitches: plain words, one meaning per term, no invented abbreviations. The vocabulary half of the sentence comes from your project instead.

Question 5

Do I need a CONTEXT.md for the skill to work?

Accepted Answer

No. The skill still works without one; you only lose the domain-vocabulary half. With one, re-pitches use the ubiquitous language you already agreed on. In repos that index multiple contexts, the skill follows CONTEXT-MAP.md to pick the right CONTEXT.md — that support landed in the 2026-08-19 update.

Question 6

Why won't the agent trigger it by itself?

Accepted Answer

Two flags lock it down: disable-model-invocation: true in the skill frontmatter and allow_implicit_invocation: false in the OpenAI binding. The design is explicit that only you know when you stopped following, so the agent never reaches for it on its own. You type /wait-what.

Question 7

Can I use it twice in a row?

Accepted Answer

Yes, and the design doc lists what 'working' looks like: the re-pitch is shorter and clearer, not shorter and blunter; it adds the premise you were missing instead of only deleting words; project nouns from CONTEXT.md replace invented ones; and a second use does not degrade into terseness.

Question 8

How is it different from grill-with-docs?

Accepted Answer

Timing. wait-what repairs the conversation you're already in — it fixes one message after the fact. grill-with-docs prevents the jargon from arriving at all by building the shared language upfront: a grilling session that runs domain-modeling as it goes and lands the agreed words in CONTEXT.md. Same vocabulary system, opposite ends of the pipeline.

Question 9

When does it not help?

Accepted Answer

When the message didn't land because the agent stated something wrong, not because it skipped context — wait-what re-explains, it does not fact-check. It also can't inject vocabulary that was never captured anywhere, so if you have no CONTEXT.md and the agent invented its own nouns, the re-pitch fixes phrasing but not the missing glossary. And if you've lost the thread of the whole session, a re-pitch of one message is the wrong tool; that's a restart or a handoff doc.

Question 10

Is the 146K install number real?

Accepted Answer

Yes — skills.sh showed wait-what at #266 with 146,544 all-time installs on 2026-08-27. Its weekly series reads 0, 0, 0, 0, 0, 34,647, 42,010, 43,703: the skill hit the public board in early August and has grown every week since. That puts it just behind writing-for-agents (151,867) and to-questionnaire (206,673), its siblings in the same conversation-repair family.

Prompting on Top AI Skills

Wait-What Skill FAQ: When Your Agent's Message Doesn't Land

Frequently Asked Questions

Why a Wait-What FAQ Exists

AI Image Generation FAQ 2026: Tools, Prompts, and Workflow Answers