シャドーイング練習: So I tried Matt's skills... - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

各モードは全画面の練習ルームで開きます。

678 文

文が短すぎたり長すぎる場合は、Editをタップして調整してください。

There is one particular GitHub repo that I've been eyeing for a long time

and just never took the time to dive into.

It's by a person that I've interacted with a bunch in the past, who if you've been around for a while,

I am sure you're familiar with.

I've often called myself the second best TypeScript YouTuber, and the reason is because of this individual, the number one, Matt Pocock.

He published a set of AI skills for real engineers that I honestly kind of dismissed initially, really.

But over time, he has continued to refine them and be realistic with them.

And what he's created is a set of small

and simple markdown files that have racked up over 200 ,000 stars on GitHub,

making it one of the 10 most starred projects on all of GitHub.

Truly insane.

All for markdown files.

I've wanted to see what the hype was about for a while here.

And I've seen people saying really good things about them, especially the grill me skill.

And this has also been a video I've planned for a while to the point where my whole

Twitch chat is freaking out Thank you.

I stayed up till 3 a .m for this.

Long overdue.

Finally, three streams ago.

Since this was put in Notion, Matt's updated the skills multiple times.

Yeah.

You guys have wanted to get my honest thoughts on these skills for a while.

In order to do that, I had to use them.

So I did.

Over the last week, I installed a bunch of the skills from Matt Pocock's skills repo,

which is full of gems that are worth talking about from his process to the actual skills themselves.

If I only covered Matt's skills, though, I'd be doing this ecosystem a disservice.

The vast majority of these skill repos are absolute slop and should be ignored.

But there is one other I'm excited about.

Pstack, created by Lauren, otherwise known as Potato, one of my old favorite

React core team members who is now at Cursor

and built a wild set of skills that are really fun and surprisingly powerful.

I have went through and audited the majority of the skills in both of these suites, pulled in my favorites, and I've been using them actively.

I cannot wait to show you everything I've learned after a real quick break for today's sponsor.

Let's be honest with ourselves.

We've moved past the era of doing one thing at a time.

And as great as this is, there's a lot of random problems it's caused.

Things like knowing where to run your code when you're working on it.

Things like making sure your CI is actually running performantly and well, and that it's doing things in parallel.

And just knowing what's going on across all of the projects that are shipping at your company.

All of this is obnoxious to do by hand and it's becoming more and more of my work.

Well, it was before I started using today's sponsor.

You've probably already heard me talk about Depot.

They're the best place for your GitHub CI and for your Docker builds, which makes everything way faster,

up to 55 times faster for a lot of real world use cases.

Their best in class runners make everything faster and their cache helps everyone on the team get started quicker too.

The cache is shared across your entire team and your CI for your Docker builds, which helps a ton with speed.

GitHub actions are also way faster to up to 10 times faster

when you use them on depot than with traditional github action ci

and if you add the depot ci runners things go even faster

because they can do actual parallelism

but that's not what i promised

i was talking about working in parallel not just building in parallel well

that's where depot's remote agent sandboxes come in making it trivial to spin up your real images

and your real projects in the cloud with your clod code

if you already have depot set up you can just run depot clod instead of clod directly

and is now running in a remote sandbox instead.

Ship faster and unblock yourself at soydiv .link slash depo.

A couple things I want to say up front before we go too deep into all of these skills

that have been provided by these different places.

The first thing, and I really want to emphasize this, you do not get anything out of blindly copying other people's setups.

You should treat these things not as a set of steps to take,

rather as a set of information to consider as you build your own systems for coding with AI

if you just copy paste my code base

and change things you're not a very good engineer but

if you use my code base as a reference point to learn

and build your skills as you craft your own code bases

then you're a good engineer the same difference applies here

if you just blindly copy my exact setup

it's like paying a bunch of money for a code template it's cringe

and bad it means you suck at engineering period so don't just blindly install all the skills here at the very least,

start with a slightly better entry point.

I see my chat realizing that

Potato was one of the builders of the React compiler.

She was one of the biggest contributors there.

She's unbelievably talented.

Yeah, PStack is going to be the underrated gem in this, and I think you guys will like it a lot.

I want to start first with how I set this all up.

I started with a prompt.

I want to figure out which

Potato PStack skills would map well to the work that I do.

I want you to audit my usage across my main machines, which were this MacBook, LeftBook, and BB1.

Compare my usage against the skills in pstack make a nice list ranking all of them by how well

they fit me

and how much they could benefit me pull history here first

then break up sub -agents to do auditing opus 5 for all

i had a bunch of opus 5 to burn at the time

so that's why i was testing this

and i linked pstack it then created this document for me where it pulled in all the skills and then ranked them.

based on how strong of a fit they were, as well as how beneficial they would be to my work, and then a little brief why.

So it immediately called out the interrogate skill, blast radius, technical writing, arena, the prove it works skill, and a few others.

I will say that this deep dive was inspired by one particular thing.

It was the interest I had in the unslop skill, because I had seen Lauren posting some of the prose that her agents wrote.

I was like, oh, that looks significantly less cringe than I'm used to Claude writing.

So I quickly grabbed this unslop skill, which is very, very simple.

There's also some fun prompting details here in the descriptions

Potato uses, because again, as I mentioned before in my like Markdown overhaul video, the descriptions role is to accurately describe everything the skill does,

because it's just fucking Markdown.

The only point of the description is to trigger the skill when it should trigger.

100

You should treat the description as the thing that gets the model

101

to do this rather than as an accurate description of everything the model can do with the skill.

102

I see it kind of similar to a YouTube video thumbnail

103

where the role of the thumbnail is to accurately encapsulate everything that the video has in it.

104

It is to get the right person to stop scrolling and click it and watch it.

105

The goal of the description is similar.

106

It's to get the right agent for the right task to decide, oh, that skill is useful.

107

I should pull that in.

108

It shouldn't be called the description.

109

It should be called the trigger.

110

All of these standards were written in the old anthropic days before anyone knew what the fuck

111

they were doing we're stuck with it it is what it

112

is anyways let's take a look at what the skills actual content is

113

so again if you're familiar with how skills work they have a name

114

and a description the models have access to all of your skills they just see the name

115

and the description and then they decide oh

116

that skill could be useful to be right now based on its description i'm going to pull it in now

117

and then what it does is it just reads this markdown file into context so when it pulls in the unslop skill, here's what it reads.

118

Edit text to remove AI patterns and add a human voice.

119

The process is simple.

120

Scan for the patterns, rewrite, preserve meaning, match intended tone, add some soul, and then self -audit.

121

Quote, what makes this obviously AI generated?

122

Then you go and fix the remaining tells.

123

Adding soul.

124

Removing patterns is half the job.

125

Sterile, voiceless writing is just as obvious. Have opinions.

126

React to facts instead of neutrally listing pros and cons.

127

Vary rhythm, short sentences, then longer ones that take their time.

128

Mix it up.

129

Acknowledge complexity.

130

Well, impressive, but also kind of unsettling beats impressive.

131

Use I when it fits.

132

First person isn't unprofessional.

133

Let some mess in.

134

Perfect structure looks machine made. Be specific.

135

Not this is concerning.

136

Rather, there's something unsettling about agents churning away at 3am.

137

Don't call me out like this.

138

Patterns to detect and fix.

139

This is one of the things LLMs are really good at.

140

If you give them an example of bad and then show them good, they will follow good very well.

141

But if you don't have certain things in those examples, they will continue to do them bad.

142

So you got to find the balance there.

143

The patterns to detect and fix.

144

The first is in content puffery.

145

So lots of these phrases, pivotal moment, testament to evolving landscape, setting the stage, cut the puffery, say what happened, name dropping,

146

listing media outlets without context, just pick one

147

and say what was said superficial dash ing phrases like highlighting

148

ensuring reflecting showcasing delete them promotional language as well vague

149

you get the idea it also calls out m dash overuse somehow

150

i still see a lot of m dashers coming out of opus the others have followed

151

this much better chat saying

152

that it's a good thing to point out how good the semantic density is here

153

i agree this is not slop this is very easy to read

154

and the info is communicated very effectively it's just well written

155

and when you have good writing in your history the models are more likely to follow it and

156

if you tell said models to follow

157

it they're much more likely to there's also a whole section here to remove chatbot phrases like i hope

158

this helps let me know

159

if found the smoking gun all those types of things no mention of load bearing sadly

160

there's a whole section about jargon this is my favorite piece, though.

161

Say what it does, not how it feels.

162

Instead of, the database stays close at hand, say something much more direct.

163

.2 SQL returns the exact string sent to the database.

164

That's so much better.

165

Ask what the sentence tells the reader to do or know, then write that.

166

If you can't restate it as a concise instruction, fact, or number, cut it.

167

One more check.

168

If the sentence could appear unchanged in another project's docs, it says nothing about this one.

169

Cut it

170

so let's see how this works

171

i will just ask with gpt 5 .6 high fast

172

you know what we'll give it something harder i'll use opus 5

173

it's going to take too long i'm going to stick with soul for this just

174

because it will be done faster what is this project help me understand it

175

and why someone would be interested in using it

176

or I could just grab like any of my other existing codex threads and you'll see the difference.

177

Here's one where I asked it to update my SwiftUI app.

178

Build 26 is uploaded to App Store Connect and processing.

179

Also completed.

180

Bullet point list of the exact things done.

181

Public distribution cannot yet be verified because the App Store Connect browser session expired.

182

183

Then I can assign build 26 to the public beta group if automatic distribution does not handle it.

184

This is so easy to read.

185

It's so much better.

186

And you'll notice when I make new prompts, one of the first things that happen is applying of the unslop skill.

187

I'm also applying the required unslop writing skill so the answer stays direct and easy to scan.

188

And then it did.

189

And now it wrote a very simple, easy to read blob of text.

190

T3 code is an open source control center for coding agents.

191

Instead of using codecs, cloud code, cursor, grok, or open code through separate terminal tools.

192

You control them through one desktop web or mobile app.

193

I'll show you the comparison and the easiest way I can think to do it.

194

I'm going to ask in chat GPT directly where I don't have my skills to look at T3 code.

195

Same rough prompt, just like said what it was specifically to make it easier.

196

Let's see how it does here without these skills.

197

The core idea is already clear.

198

T3 code is not another coding model and not mainly an editor.

199

God, I didn't realize how quickly I adjusted to unslopped text.

200

And now when I read the slop text, it hurts me.

201

It's actually insane how much of a difference I can already see just from that.

202

Again, T3 codes and open source control center for coding agents.

203

The core idea is already clear.

204

What using it looks like.

205

A normal workflow is run the command, open a project, start a thread, ask it to inspect or change the project, review messages and code changes,

206

continue from your desktop browser

207

or phone t3 code also manages terminal terminals agent sessions get diffs

208

and checkpoints each turn can have a checkpoint which lets you inspect

209

or revert what the agent changed short version t3 codes across flower control center

210

this is the like quote from the docs

211

it turns these providers from separate terminal programs into one coherent agent workspace m dash with desktop web ios

212

and android clients there's like t3 code is and t3 code is not the strongest pitch is not

213

Chat with AI in nicer windows.

214

It is operating several autonomous developers without living in several terminal tabs.

215

Remote agents become practical.

216

One interface across providers.

217

It's designed for parallel work.

218

The especially compelling use case for your setup.

219

Versus why someone would use it.

220

One interface works with several providers.

221

Threads persist, so it works easier to resume.

222

You can control agents on another computer from your phone or browser.

223

Your files

224

and provider credentials remain on the machine during running the server

225

it's insane these both came from the same model for adding one markdown file it's

226

so much more readable i have been enjoying coding

227

so much more to install this one stupid skill thank you potato

228

and thank you pstack for providing me this wonderful unslop skill hopefully you can now understand why i

229

so quickly went down this rabbit hole it's because I was really impressed with that skill once I installed it.

230

So I then went through all the pstack skills to see which other ones might fit me too.

231

And as I was doing this, I had the realization that I'm sure you guys could have predicted

232

that I also should probably look at the Matt Pocock skills.

233

So I linked that as well

234

and ended up getting another document of all of Matt's skills and which ones fit my use case as well.

235

Now it's time to dive in to all of these skills and what value they can bring you.

236

I've been focused on pstack for a bit too long.

237

So let's hop over to Matt's skills instead.

238

The main directory that we care about is skills slash skills slash engineering.

239

There's ask Matt, code review, code base design, diagnosing bugs, domain modeling, grill with docs, which is an update to his grill me one,

240

implement, improve code base architecture, prototype research, resolving merge conflicts, set up Matt Pocock skills as a skill itself.

241

TDD, to spec, to tickets, triage, wayfinder, and wizard.

242

I want to start by playing with a few of these.

243

I'm going to grab the improved code -based architecture in Grill with Docs.

244

If you're ever curious about a skill and how it behaves, the easiest thing to do isn't to install it and then hope for the best.

245

It's actually quite simple.

246

You can usually just copy paste it.

247

In this case, this skill is calling for other skills.

248

It's actually hilarious how simple the grill with docs is it's a relentless interview to sharpen a planner design

249

which also creates docs adrs

250

and glossary as we go he also has disabled model invocation on for a lot of his skills

251

which means the model won't enable it itself you have to manually pull it in with a slash command

252

or a dollar sign command or something like that

253

which i think is a very good call

254

because it was how some of his skills work

255

but some of them that isn't necessarily the case like when I'm going through the diagnosing bugs skill.

256

This one I probably shouldn't have to manually pull in

257

and sure as hell he doesn't have this as a manually pull in skill.

258

This is a skill that will invocate itself when it's helping debug issues.

259

And I've had this skill fire

260

and help with a lot of my debugging stuff over the last few days and it's been pretty solid.

261

It seems like my agents find the root cause and can communicate what's wrong much more effectively.

262

So let's look at the new grill me and grilling skills because these are the ones that are referenced there.

263

Here is the grilling skill.

264

Grill the user relentlessly about a plan, decision, or idea.

265

Use when the user wants to stress test their thinking or use any grill trigger phrases.

266

Interview the user relentlessly until you reach a shared understanding.

267

Map this as a design tree.

268

Every decision branches into the decisions that hang off it.

269

Work the tree in rounds.

270

The frontier is every decision whose prereqs are already settled.

271

God, why is he using em dashes in his skills?

272

Matt, Matt, how many are on this? there's nine m dashes in this page matt

273

i was wondering why i was getting m dashes again

274

it might be matt skills there's at least one per paragraph for most paragraphs oh

275

this hurts me i'm still gonna try it

276

and i told you guys it's an easy way to try skills

277

you copy the text go to your agent you do whatever

278

you want to do then you paste it i'm gonna ask it to grill not me and ask it to grill LakeBed.

279

LakeBed, paste, pick machine.

280

I want you to grill me about LakeBed.

281

The project itself and its goals, everything in the agents, MD, Claude, MD, et cetera.

282

I want to make sure we are on the same page about what we are building.

283

I want to use Matt Pocock's grill skill.

284

don't have it installed i will paste the text below

285

and now i have successfully used the skill without installing anything by just copy pasting the text obviously

286

this will not work with skills that have other things like scripts associated

287

and whatnot but you can copy the script over it's fine

288

or just tell the model about it it can go find the script

289

and download it it's fine this is the easiest way to test out a text -only skill, literally just copy -paste the text.

290

And now we have this Opus 5 run on my remote

291

Linux box running, and it will grill the fuck out of me momentarily.

292

It took almost two minutes, but it has its first pass here.

293

CloudMD is badly stale.

294

The repo contradicts its own v0 rule, and there are two competing execution models in the tree.

295

question one who is the user

296

and who pays as well as example answers we could give question two what is the actual next milestone

297

because there's a mess of like launch prep files html files with like random plans the to -do md

298

that's an organized from a pr cleanup program not as launch q3 is the capsule format a spec

299

or just our implementation what is the ceiling of one capsule.

300

These are actually good questions.

301

And now I'm curious, do we get similarly good questions if I ask a different model?

302

Let's switch over to Sol with the exact same thing.

303

Oh, I already had the grilling skill installed.

304

Yeah, I did carry that over.

305

God, the 5 -6 autism is so real.

306

The grilling skill requires me to find repo facts instead of asking you.

307

I just realized I use this on low and slow.

308

I should be using this on high and fast.

309

Get better answers faster.

310

I will stop that because I don't care anymore.

311

Oh, wow.

312

It finished before the stop even persisted.

313

That's hilarious.

314

What is Lakebed?

315

Who is the primary customer?

316

What does full autonomy mean?

317

And what is a capsule?

318

All good questions.

319

Let's see how the Hi version does.

320

Oh, and Hi, it spun up some sub -agents in the background to do more hunting to get more data for future questions.

321

Start with the same what's Lakebed?

322

Who is the primary user?

323

All with like quick answers, which makes it really easy to go through just like 1A,

324

2b 3a etc again like the question asking tools in a lot of these harnesses

325

and agents just aren't good enough to have the like question asking

326

you why come up this is a lot quicker to just blast through

327

so one i'm gonna say it's the autonomous app platform two i'm gonna say these two are the users

328

that matter three doesn't give me a simple answer

329

but i can just tell it what i don't want which is agents who do everything other than payment details.

330

Four, how much should Lakebed own?

331

Own everything as needed.

332

Five, which rule wins?

333

The current guidance contains principles that will conflict.

334

I want simple where possible, but also rewrite where beneficial.

335

Six, what's the product boundary?

336

Should Lakebed remain a real open and useful system while the host of cloud platform is the production path?

337

Local is only dev sim.

338

That's nice and easy.

339

And I'm already realizing that I need to clean up this project based on the simple, quick start of this grilling section.

340

Okay, I'm seeing the light.

341

This is useful.

342

I should have tried this before.

343

Let's keep going through these skills.

344

Teach is one I've heard really good things about.

345

This one is also disabled model invocation.

346

So you have to trigger this with slash teach.

347

The user's asked you to teach them something.

348

This is a stateful request.

349

They intend to learn the topic over multiple sessions.

350

Interesting.

351

Treat the current directory as a teaching workspace.

352

The state of their learning is captured in this directory in several files.

353

Very interesting.

354

So you're supposed to use the teaching skill in its own folder.

355

Huh.

356

I want to read more from the top level.

357

I feel like I am not going through this the right way.

358

Interesting.

359

One of the skills is Ask Matt, which is specifically meant to help you

360

You figure out what skills or flows fit your situation.

361

There's grill with docs, as we talked about before.

362

Triage, which helps move issues through a state machine of triage roles.

363

Improve code -based architecture.

364

This is the one I actually wanted to try.

365

A lot of these are self -referential,

366

where the code -based architecture one will say to run the code -based design skill in order to get the vocabulary.

367

Present candidates as an HTML report.

368

Nice, if somebody likes HTML skills,

369

I like this a lot context md vocabulary for the domain then it calls out grilling as well as domain modeling

370

wayfinder is a fun one too for planning huge chunks of work

371

i also really like the separation in the docs between user invoked

372

and model invoked because these are very different like these are effectively plugins

373

that you call

374

when you want them yourself model invoked is more steering the model to do different things in different ways

375

and these ones are very useful

376

Like I mentioned before, the diagnosing bugs one has helped me a ton already.

377

Wizard seems really cool too.

378

To walk human through steps only they can perform.

379

This is super cool when like an agent can't access some dashboard it needs to

380

or there's a pseudo call that it can't run.

381

Wizard is a guide to help the agent set things up

382

so that you can just run the one script and then handle things from there.

383

Do I have this one?

384

Because if I don't, I'm going to go at it now.

385

I don't.

386

I guess here I'll give you the spoiler of how I actually set up all of these things.

387

I did all of this in a thread in T3 code in a repo

388

that I already made called fleet where I manage all of my skill files and things.

389

So in here, I'm going to do this specifically.

390

I'm going to ask, did we set up the wizard skill from Matt's repo?

391

If not, set it up now note

392

that we have one more machine in the fleet to deal with well

393

because since i last worked on this i added my new mac mini

394

so i want to just give it a quick little hint

395

that the new mac mini exists so that it will touch

396

that as well apparently

397

i should have been reading the docs the whole time my

398

bad yeah the getting started is the setup skills command

399

which helps you like set up issue tracking and whatnot in a way

400

that the skills will handle

401

it the main flow is to start with grill with docs get interviewed about a plan

402

and record the decisions then to spec turn into a spec then two tickets to cut

403

that into tickets for jira linear github issues whatever you want to use implement for actually building it

404

and then code review for reviewing

405

it after shaping is fun for turning vague ideas into real executable plans wayfinder helps

406

you chart a large effort as a map of decisions

407

and settle them prototype is for actually like making designs

408

and mocking things to see what you really want before you ship it just deleteable code

409

and the research skill for getting cited answers from primary sources i don't know

410

if i would ever need

411

that i feel like the agents are pretty good at verifying their claims

412

if you just ask then we have improved code -based architecture

413

this one seems fun diagnosing bugs resolving merge conflicts triage

414

and wizard i am excited to try out wizard for activity

415

skills like grill me handoff to questionnaire teach wait what

416

and writing for agents then the reference skills like code -based design vocabulary for deep

417

or designing deep modules covering the words a project uses

418

and writing them down domain modeling i think is a really helpful one

419

i've been in like writing down my own grammars

420

and like glossaries with agents in every project on like what terms

421

we use to do different things the idea of being a skill does make sense to me

422

i am increasingly tempted to just install all of them

423

but i'm very happy with my current setup and i don't want to muddy it too much more.

424

So I'll continue doing what I normally do,

425

which is just telling my agent to rip the parts I want and keep them in my fleet directory.

426

This little fleet repo has saved my ass so many times.

427

Well, that is running.

428

Let's see the follow -up here.

429

Okay.

430

We got a bunch more questions.

431

Which form of simplicity matters most?

432

Oh, this is brutal because I want all of these types of simplicity.

433

God, this gets deep.

434

I don't want to answer these questions.

435

That's the point, isn't it?

436

I really don't have to answer this because my answer is all three of these.

437

I want them all.

438

Okay, I said I will begrudgingly pick A even though I want them all.

439

What apps must LakeBed support?

440

A and B, not C.

441

How do the two primary users divide control?

442

If the developer and agent are both primary, what does each one own?

443

So like who chooses architecture, data models, deploy timing, et cetera?

444

Dev chooses nothing but app functionality.

445

Cool, that's it.

446

That is answered.

447

10. Does everything include irreversible actions?

448

Authorized agents be able to delete production data.

449

Yes, they should be able to do whatever.

450

Where is the payment boundary?

451

Will be sub plus monthly limits.

452

Agents can use them as they please.

453

12. What must the local simulation preserve be?

454

Cool.

455

This is great.

456

I'm actually disagreeing with the recommended answers pretty often too.

457

fuck this is actually going to help me I'm annoyed now this skill is on all my machines cool

458

opus caught all the changes to my systems

459

I know it's silly but like all this text is

460

so much more readable than older agentic stuff

461

I was doing in like the same project with the same code base and the same agents and models and everything thing.

462

I asked a question and it gave me a direct answer.

463

No, wizard was never set up.

464

It ranked seventh in the previous report, but you only pick seven skills and it wasn't among them.

465

It's installed now.

466

Whereas like if I scroll back far enough to things from before I had these set up, just like any old thread,

467

here's a recent changes summary.

468

Here's a ready to share write up.

469

I treated landed as changes that reach main.

470

T3 code the last three days.

471

Oh no, this is it trying to like match my tone.

472

Release publication now uses the job scoped github token instead of

473

competing for the right the shared release app api quota fixing repeated nightly failures oh god the m dash hell

474

i said push straight to his branch done the follow -up

475

commit is now on his pr branch directly m dash 5561 shows it's the fifth commit

476

and the review bots are rerunning against the new head his

477

pr is still based on one merge behind the thread pagination pr landed since it's still mergeable per github

478

so no action needed rebasing his branch under him felt more

479

intrusive than the situation called for worth watching whether the effect service convention check goes green now m dash

480

that was the hard blocker in my context reference approach is a lighter lighter remedy oh god

481

i can't even like read this

482

and i'm trying to like read it out loud for you guys in a video it's

483

so much worse god life post slop is unbelievably cool oh god

484

i got another like 10 plus questions for me i'm gonna do this off stream

485

but i'm actually going to do it

486

because these are good questions this does remind me

487

that i have a branch i was working on in t3 code

488

that i'm i have new ideas for now too i was working on a skill manager

489

feature in t3 code yeah this cross computer skill manager page

490

that i was working on to make

491

it easy to just manage skills on your machines through like

492

one layer what i want is now is the ability to

493

disable them en masse like to have groups of skills like the p stack skills

494

and the matt pocock skills and be able to turn them all on

495

or off or just turn on the ones i want oh

496

that would be

497

so nice yep the wizard skill has now appeared on all of my machines yay i don't have anything

498

that needs wizard right now but i am very excited to try

499

that one it makes a ton of sense to me

500

i will say most of these skills feel more like prescribing like mental workflows

501

and like giving you an easy entry point to try them out,

502

especially so like the main flow section here with the grill with docs to spec to take it implement code review.

503

I don't necessarily want this much prescription on how I go step by step, like I've been building my own workflows,

504

and they don't map quite as well to traditional stuff.

505

There's a ton of good things that I'm grabbing from here, like the wizard skill, the grilling skill is clearly super, super useful.

506

I'm going to play more with the code -based design and the follow -up cleanup ones.

507

I have heard

508

Wait What is really good, as well as the Writing for Agents.

509

The Writing for Agents one I've actually been using a bunch.

510

Writing Great Skills was turned into Writing for Agents.

511

This is useful because agents are really bad at writing instructions to other agents.

512

If you let your agents write skills for you, write markdown for you, write prompts to subagents for you, all those types of things,

513

it can do a very bad job.

514

So giving it better instructions on how to do it right sounds very compelling to me.

515

Wait, what's another one that I'm really excited about?

516

It pulls in Simplify Technical English, which a dedicated video on this bit coming in the near future.

517

It's a simple skill that's user invoked in three lines long.

518

The point is to get a simpler, easier to digest description of something when you get a response that makes no sense.

519

He does call out that this only repairs one message.

520

It doesn't prevent the next sloppy one.

521

The solution

522

and the cure for this type of jargon is a shared language built up front using the grill me with docs skill.

523

Reach for wait what when you don't have that setup done yet.

524

I am very excited for wizard I'm going to be using

525

that a ton I can think of like four things

526

I should have used it for yesterday

527

I really like how he's documenting all of this stuff like this is a good docs page

528

there's a lot to learn from here in the best sense here we are wait what super simple

529

and again it disable model invocation it means

530

you just type in the one skill invocation with the slash command dollar sign whatever wait not m dash ah

531

i almost want to make a dm dashed matt pocock skills fork where it's just the exact same repo

532

but all the m dashes removed the description stop

533

that last message did not land m dash repitch it god

534

i i almost want to see how many m dashes are in this project yeah

535

it's just one regex remove it all i know this is one

536

i was actually really excited about the the writing for agents oh

537

this one's long good it should be reference for writing any document that an agent consumes a skill an agent or ClaudeMD,

538

a doc reached by a pointer.

539

I've mostly been using this for prompting sub agents, and it's been very helpful there.

540

Too many MDashes still.

541

When the document you're writing is a skill, read the skill mechanics MD file for front matter invocation choice and router skills.

542

Context pointers.

543

Context pointers are reference held in the agent's context

544

that names some out of context material and encodes the condition for reaching it.

545

The skills description is one.

546

Align an agent MD naming a doc is the same object.

547

The pointer's wording, not its target, decides when an agent reaches the material

548

and how reliably a must -have target behind a weekly worded pointer is a variance bug

549

this is unreadable i mean i'm not surprised

550

but god damn ow versus unslop just a chunk of it no no

551

i don't trust these anymore does it just seem markdown as ai generated is

552

that my problem here cool i passed 100 human written yeah very likely all of these were written by ai

553

which is annoying because i have had not great luck with ai written stuff i'm amazed

554

that the unslop one is allegedly ai though because

555

that skill read very easily and was very dense

556

and well written i will say regardless of

557

if they're both ai generated i find p stack writing to be a lot more readable

558

and also the behaviors from these skills to be a lot

559

more applicable for my day -to -day the arena one has been super fun fan out

560

and parallel attempts at the same task read every candidate

561

end -to -end pick the strongest as the base graph the

562

best ideas from the others into it verify the synthesized results

563

this has been a very fun skill for those of us who are token burners

564

because we have a bunch of usage to get through started

565

to do list with one entry per phase before launching anything the arena runs autonomously

566

and the list keeps phases from silently disappearing frame fan out cross judge pick graft verify

567

then descriptions of each phase surprisingly little text for how much

568

this does like this one skill has cost me hundreds of dollars of inference

569

because it just does the same thing multiple times this write -up is very much tied to cursor specifically

570

which is the biggest issue with p stack i honestly would be pumped

571

if somebody like cloned all the p stack skills in a generic not cursor specific way

572

because there's so much gold in here

573

but a handful of these are just a little too cursory

574

she did also pull the bro skill from uh i believe it was originally dylan mulroy the effect

575

and functional programming javascript guy restate your last message stop using jargon

576

and speak coherently state it more simply

577

and concisely like one human talking to another i think

578

this is basically identical to wait what yep pretty much identical also one of the few

579

that potato has is disable model invocation, which is interesting.

580

She normally just lets the model do its thing.

581

Blast Radius is awesome, too.

582

I've been using this one a ton.

583

Find what a change breaks somewhere else before it ships.

584

Use Blast Radius of X.

585

What could this break?

586

Or reviewing a small diff that you don't trust.

587

Compare your companion to how and why.

588

How tells you what the code does.

589

Why tells you why it's shaped that way.

590

Blast Radius tells you what it breaks somewhere else.

591

I know this calls out you can't trust your own write -up, specifically saying that the history of this thread is not trustworthy

592

so don't hand back the write -up find the one

593

or two facts the whole thing depends on

594

and prove them by running code this one is great

595

and has cop a couple things that would have been miserable

596

if i didn't have it it also calls out

597

that it should write the response through unslop to make sure

598

that it's not a mess this is great i didn't want this to be like a one

599

or the other thing where it's like pstack versus pocock skills i didn't intend for it to be

600

that at all

601

but i am much more philosophically aligned with what potato is cooking over here oh yeah

602

she also has the show me your work skill

603

which is really cool

604

if you ever want to understand why an agent did something

605

to like keep track of what's going on keep a reviewable decision trail for long running

606

and unattended work a tsv log with one row per decision what why evidence

607

and result local by default commit it

608

when a reviewer needs the trail to trust the result use for slash show me your work, autonomous or multi -phase runs,

609

or work a human reviews after stepping away.

610

Gives it a format on how to log its work

611

and its decision -making process all in a simple TSV format so that it's easy to read and do in other things.

612

One row is one decision or checkpoint.

613

If it doesn't fit one line, the decision isn't crisp yet.

614

Append only.

615

A wrong call gets a new row that supersedes it.

616

Prefer evidence produced by committed scripts over handmade one -offs so a reviewer can rerun it.

617

Interesting.

618

This one seems really cool, actually.

619

I'm going to play with this more later.

620

This video is supposed to be an overview, but what you're getting instead is my actual process I did pretty much exactly everything you're seeing here

621

but with slightly less depth a few days ago

622

and that's how I found the cool skills that I did find

623

and this is what you should be doing as well you shouldn't just blindly install a bunch of shit

624

when you could take the time to read the things

625

that map up to the work you do and how you do things

626

and pull over the parts that you actually want

627

and it all should start with something small like grab the one you like.

628

I think everyone should have unswap at this point.

629

This skill has fundamentally changed my willingness to read the things that my agents say to me.

630

It's been great.

631

God, this video could be hours long or it could be a 45 second short.

632

It could be a lot of different things.

633

It isn't quite what I expected it or wanted it to be, but I hope you can get some good takeaways from it still.

634

The first thing you should do is you should grab your agent

635

and you should open it up and tell it about pstack

636

and tell it about matpocock skills and ask it to look through them

637

and then look through your histories

638

and figure out what makes sense based on what you do maybe you should install the unslop skill before

639

that though because then you'll actually want to read the responses you get afterwards you should look through

640

that you should read the markdown for the skills you're installing before you install them then you should decide

641

which ones you want you can tell your agent to set it up you can do a single repo

642

that represents all your skills like I do.

643

I think it's really powerful.

644

You have a lot of options here for how you choose to manage all of this.

645

But you should be managing it.

646

You shouldn't be blindly copying someone else's setup.

647

You shouldn't be running commands that install 500 skills you don't know anything about.

648

You should take this opportunity to refine your toolbox based on what you see working for others

649

and what you try setting up yourself.

650

You should, based on your own history and your own usage, make subtle adjustments and changes.

651

And you shouldn't be scared of editing these files.

652

I've noticed a lot of people are scared of opening up their clod files

653

and like the dot clod and the dot agents directory on their machines.

654

That's silly.

655

And you have to get over it because there is so much cool shit you can do.

656

The moments you start digging in and playing with these things, you got to be willing to edit those files.

657

I joke now that the only time I open my code editor is to edit markdown files, but like it's barely a joke at this point.

658

It's markdown files and environment variables at this point.

659

And it has helped a ton.

660

It's one of the biggest shifts that has allowed me to do way more parallel agent work.

661

It's a huge part of why I was able to ship so many PRs with so little breakage over the last weekend.

662

The way I work is really different now.

663

And it's largely because I took the time to go deeper in these directories

664

and pull together skills for myself and others that make sense for the work that I do.

665

And I hope this inspires you to go do the same.

666

Explore these collections of skills people have published.

667

Don't blindly install them.

668

Poke around at them.

669

Try them out.

670

Pull the parts you like.

671

Ignore the parts you don't.

672

Play.

673

I think I've said all I have to here.

674

There's some really good stuff in Matt's skills and even more good stuff in Potato's skills.

675

Check both out and maybe make some of your own too.

676

I'm curious how this affects your workflows.

677

Let me know.

678

And until next time, peace, nerds.

アプリをダウンロード

流暢に話すために必要なすべて

AI発音採点文ごとに採点

IPA練習すべての音をマスター

単語学習自分の単語帳を作る

単語ゲーム遊びながら覚える

QRをスキャン

Download on theApp Store Get it onGoogle Play

同じチャンネル

PirateSoftware is right, this needs to stop

Daily ConversationWorking Adults

OpenAI fights back

Daily ConversationWorking Adults

FOR YOU

このレッスンの語彙とスピーキングのポイント

このスピーキングレッスンは、動画「So I tried Matt's skills..」を教材にしています。繰り返し出てくる語は次のとおりです：skill, agent, code。この動画には、シャドーイング用の文が678文、単語が7604語あります。音声の長さは38:21です。話す速さは速く、1分あたり約198語です。音のつながりや弱く発音される音が多くなります。英語の頻出3,000語に含まれる単語は80%だけなので、語彙は難しめです。

この動画の重要語彙

動画に出てくる覚えておきたい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
skill 名詞	/skɪl/	腕, 技
grill 名詞	/ɡɹɪl/	グリル
dash 名詞	/ˈdæʃ/	長点
install 動詞	/ɪnˈstɔl/	導入する, インストールする
wizard 名詞	/ˈwɪzɚd/	魔男, 魔法使い
thread 名詞	/ˈθɹɛd/	糸
potato 名詞	/pəˈteɪ.təʊ/	じゃがいも, 馬鈴薯
architecture 名詞	/ˈɑː.kɪˌtɛk.t͡ʃə/	建築, 建築学
parallel 形容詞	/ˈpæ.ɹəˌlɛl/	平行の
pasta 名詞	/ˈpɑ.stə/	パスタ
script 名詞	/skɹɪpt/	文書, 文章
blast 名詞	/blæst/	疾風, 突風
trigger 名詞	/ˈtɹɪɡɚ/	引き金, 鉷
depot 名詞	/ˈdiːpoʊ/	倉庫
directory 名詞	/dɪˈɹɛktəɹi/	一覧, 名鑑

動画に出てくる句動詞

単語	意味
clean up 動詞	掃除する
come up 動詞	上がる, 登る
go down 動詞	降りる, 降下する
point out 動詞	指す, 指し示す
pull over 動詞	道の脇に寄る, 道の脇に寄せる
try out 動詞	試す, 試してみる
write down 動詞	書き留める, 記す

繰り返し練習したい文

動画に出てくる短く完結した文です。日常会話でそのまま使えます。

I am sure you're familiar with.
Let's be honest with ourselves.
Don't call me out like this.
Do I have this one?
Wait, what's another one that I'm really excited about?

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
現在完了進行形 have/has been + -ing — 以前に始まり今も続いている動作	I've been eyeing · I've been using · have been enjoying
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	I've interacted · I've often called · has continued
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	was put · should be ignored · is shared
関係詞節 who / which + 節 — 人や物について情報を加える	builds, which makes · machines, which were · skill, which is

注意したい発音

話し手はI'm, don't, I'veなど、短縮形や弱形を135回使っています。聞こえたとおりの短い形で発音しましょう。

「th」の音: thread /ˈθɹɛd/, thumbnail /ˈθʌm.neɪl/
「sh」と「zh」の音: dash /ˈdæʃ/, speculation /ˌspɛk.jəˈleɪ.ʃən/, instruction /ɪnˈstɹʌkʃən/, beneficial /ˌbɛn.ɪˈfɪʃ.əl/, cache /kæʃ/
長い単語（アクセントの位置に注意）: architecture /ˈɑː.kɪˌtɛk.t͡ʃə/, speculation /ˌspɛk.jəˈleɪ.ʃən/, directory /dɪˈɹɛktəɹi/, autonomous /ɔˈtɑnəməs/, vocabulary /vəˈkæb.jə.lə.ɹi/

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: grill /ɡɹɪl/, parallel /ˈpæ.ɹəˌlɛl/, vocabulary /vəˈkæb.jə.lə.ɹi/, necessarily /ˌnɛs.əˈsɛɹ.ə.li/, trail /tɹeɪl/
/v/ — /b/ にならないように、上の歯を下唇に当てる: versus /ˈvɝsəs/, dive /daɪv/, provider /pɹəˈvaɪ.də(ɹ)/, vocabulary /vəˈkæb.jə.lə.ɹi/, impressive /ɪmˈpɹɛsɪv/
/f/ — 「フ」ではなく、上の歯と下唇で出す: fleet /ˈfliːt/, phrase /ˈfɹeɪ̯z/, format /ˈfɔːɹ.mæt/, helpful /ˈhɛlp.fl̩/, interface /ˈɪn(t)ɚˌfeɪs/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。skill, grill, dashなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: So I tried Matt's skills... - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

アプリをダウンロード

同じチャンネル

PirateSoftware is right, this needs to stop

OpenAI fights back

おすすめ動画

It’s not just you

Opus 5.5 Just Changed Video Editing Forever (free skills)

mattpocock/skills: A complete AI Coding workflow, end-to-end

Fluent Forever

My 4.9-Star iOS App Is Actually a Website: Here’s How I Built It

How I Learned Procedural Generation

I Tested The Viral Reverse Engineering Repo

9 Insane Claude Mods I Already Can't live Without

Go’s Big Mistake

GitHub Copilot now controls your command line...

The hottest programming skill in 2026

The Only AI Tools You Need (12-Minute Guide)

このレッスンの語彙とスピーキングのポイント

この動画の重要語彙

動画に出てくる句動詞

繰り返し練習したい文

この動画の文法

注意したい発音

この動画での練習方法

シャドーイングとは？英語上達に効果的な理由