mattpocock/skills v1.2 is here! Brings /wait-what (for smacking down verbose models), /writing-for-agents, and a huge change to /grill-me Here's the breakdown:

Here are the brand-new docs!

Nice update Matt. 👏 I often use /writing-great-skills to audit and simplify the skills we have, or whenever somebody creates a new skill. Would be great if you include this feature in the skill. On the same topic AGENTS.md audit would be great too. Cheers

Very nice website/ resources Matt! Love it! Summer holidays lectures coming 2 weeks in France!

Thank you Matt ! I know many people are relying on AI heavily for writing Skills, and they are bloated with meaningless fluff that actually degrades performance. These skills fill the gap and would greatly help everyone to write skills for Agents.

I really don’t like the new multi questions grilling, when doing solution design as a team it just make each question less precise and also make it harder to reason at multiple people on one precise topic instead of 6 at same time, can we make it configurable using agent/*.md ?

I have one request rename wait-what to say-what

Tried /prototype on a non-ui task today and was pleasantly surprised how well that worked out. Thanks for relentlessly encoding all this knowledge and best practices into skills for us.

naming it /wait-what is half the UX. everyone knows exactly when to use it.

I need your voice as my AI Voice agent. How do we make that happen. (no homo..LOL) Love your content, thanks for sharing!

I’ve used wayfinder for the first time and this is the best change to my workflow since discovering your grill-me skill months ago. Thank you so much.

Wizard is also looks very promising 😀I don’t really like aws, and if it can ease the pain, that can be really great

Gold. I’ve been struggling with documentation. Gosh. Bless you Matt.

The new grill me workflow is much worse. The questions and possible answers are mixed in together. Needs guidance to improve the formatting of the answers to stand out separate from the question.

echo "When grilling, ask one question at a time" >> ~/.claude/CLAUDE.md

I don't mind all the questions, it's just the formatting that needs work.

What would improve it?

There's a template in there, have a play to see if you find something you like better

The question followed by an indented set of a, b, c bullet points which are the possible answers. Right now it puts all the answers inline which is really not a great reading experience. Format it more like the "recommended answer" component.

Hey @mattpocockuk. I'm curious as to your hygiene around installing skills. Do you recommend installing these per-project as needed, or are you an advocate for install in ~/.agents?

For solo devs, global install is fine. In teams you should use them at the project level so that you can all contribute to them.

Beating this horse to death

hm all skills appear in claude code but not /wait-what how the heck, plugin is updated to latest as well

Many thanks for /wait-what skill. Finally, felt normal while reading the response of opus 5. @AnthropicAI You should give this man a medal 🏅for this skill. Much respect🫡

太好用了，我喜欢

Super useful skills, Matt. Thanks, sir.

emojis 🥹❤️ people too harsh on them

/wait-what 这个名字一看就很解压

finally, a /wait-what command for when my own models start hallucinating.

/wait-what is 🔥

AI 模型的 benchmark 越来越卷，但落地时发现最大的问题是场景理解。技术跑分和真实需求之间，隔着一条叫做'产品思维'的鸿沟。大多数 AI 创业公司，死在拿着锤子找钉子的路上。

I added /wait-what to my custom prompt setup last week. Actually forces the model to trace the state instead of hallucinating a fix.

To be honest I don’t like the grouped question in this new version. It’s so hard to read the question and option because of the formatting. Is there a way to go back to the older option?

how do i update it? do i have to re install it from scratch?

npx skills update

the verbosity problem rarely gets named until someone's paying per token or grepping through a log that's 80% hedging. curious how „/wait-what” enforces it - a system prompt instruction, or something that actually truncates/rejects the response if it runs long?

The /wait-what idea is such a clean fix for model rambling — curious how aggressive the defaults are

Since Opus 5, I’ve constantly struggled to understand the model’s answers. I’ve honestly wondered if they’re written for other high-level models rather than humans. I’ve been asking for clarification almost every time for two weeks, so yeah, that wait what skill might help 😅

Love this. We've been going down the same path with agent-specific workflows.

The batch grilling with emojis is amazing dude. I just did a 4 round grill session in 3 minutes instead of 20😂.

Seeing /writing-for-agents added fits agent prompt needs

@mattpocockuk What's your official stance on stuff like Beads? Agnostic? Anti? Pro?

Seems fine! But I prefer my issue tracker state to be remote so I can easily share it

I think beads *does* store tracker state directly in Github if you want it? At least last I checked?

In that case, great!

Thanks, really appreciate your time