Lauren Begins Guide to Pstack Agent Engineering Workflow
Engineer Lauren, known as @poteto, begins a multi-part guide to pstack, her set of skills for rigorous agent-assisted engineering, claiming it enabled about 2,000 PRs a month. Part one centers on verification skills that let agents check their own work.
Original post Β· 12 min read
I'm writing a guide to pstack! Here's part one.
X ArticleThe Complete Guide to pstack Pt. 1
In this series of posts, I'm going to show you how I use pstack, my personal set of skills for doing rigorous engineering work. It's allowed me to ship 2,000 PRs a month to production with high confidence.
Personally, I have never put much emphasis into how many lines of code or how many PRs I was landing. Before agents, no one cared, and rightfully so, as raw productivity did not always equate to quality or a visible outcome for users. It was simply a vanity metric.
But I've discovered through the course of building pstack that volume does matter, especially when you are able to maintain or even increase the level of quality of the product with agents. For example, I started working on Grok @Bot about 2 months ago, when it was still in its early days and the codebase was fresh but starting to grow. Despite the team growing and now landing hundreds of PRs a day into the Grok @Bot codebase, pstack has allowed me to keep the quality of the code high for everyone as I constantly monitor code, refactor, add new lints and checks, and also work on features.
Being Grok @Bot's gardener and maintainer is something I was only able to do through pstack. Our early momentum after building the prototype was very high and many people were joining the team. I had a critical moment of opportunity to refactor the whole codebase, while it was being built and extended and with no downtime, into something with strong foundations. A codebase with high quality that scales no matter how many engineers (and most importantly, non-engineers) contribute to it. All of this work requires me to refactor and improve the foundations of Grok Bot as it's being built, and you can only do that when the foundations can keep up with the number of contributions.
The proof is in Grok @Bot itself. Over the next few weeks, I'll tell you everything you need to know to be able to build and maintain a high quality app using pstack.
Part 1 β Verification is all you need
The most critical skill to have in your toolbox is a high quality verification skill. This skill is so important to have and maintain that I think of it more like critical infrastructure rather than "just" a skill. A good one will amplify the output of your whole team, including non-engineers. Done well, you will 100-1000x your whole team's output.
If you're not familiar with the term, verification means that an agent can verify its own work. It can keep going until it succeeds at its task, because it can now close the loop without you being the bottleneck. If you're interested to know more of the story of how I created my first verification skill for Cursor, check out my previous post Loops You Can Trust.
Let's build a verification skill together
To start, install pstack and then run /create-verification-skill. I also recommend adding Dr Eggbot, my bot that helps you create high quality bots, to your roster. Dr Eggbot ships with pstack. Itβll teach coding bots how to use it, and it can also make non-coding bots with the same rigor.
You can ask Dr Eggbot to create an engineer bot for you that you can then ask to run /create-verification-skill and set up a daily routine to run /maintain-verification-skill.
While that runs, let's walk through what the skill does and how it makes a high quality verification skill for you.
I distilled all of our verification skills that we use to build Grok @Bot and Cursor into this skill as a sort of meta-skill. It teaches your agent how to create a high quality one for your own app.
Now this is where choice of tech stack is important. If you're building an app in Electron or for the web for example, you can take advantage of the rich debugging tools available for the JS ecosystem. For example, the Chrome DevTools Protocol (CDP) allows you to use the same tooling available in your browser's developer tools. Or if you're building an iOS app, making use of the simulator.
You ideally want the ability to interact with your app, debug it, take perf traces, and any other debugging and development tooling that you might typically use if you were developing the app by hand. If you don't have a rich runtime to make use of, you may need to ask your agent to create tools for you (eg using lldb, or a custom package that runs as a sidecar in dev environments), or just make use of what you have available.
I personally feel that agentic verification is so important that I would unironically suggest building your own rich debugging tools, or even choosing a different tech stack, in order to have unfair advantages and extreme productivity in building software. As I mentioned earlier, giving agents the ability to verify their own work unlocks everyone in your organization to be able to contribute and validate that their changes actually work. The harder your tech stack is to debug and control, the more difficult it will be to use agents productively.
Make it Reproducible
In pstack, we have a principle called "Build the Lever". What this means in the context of cβ¦ continue on X β
Personally, I have never put much emphasis into how many lines of code or how many PRs I was landing. Before agents, no one cared, and rightfully so, as raw productivity did not always equate to quality or a visible outcome for users. It was simply a vanity metric.
But I've discovered through the course of building pstack that volume does matter, especially when you are able to maintain or even increase the level of quality of the product with agents. For example, I started working on Grok @Bot about 2 months ago, when it was still in its early days and the codebase was fresh but starting to grow. Despite the team growing and now landing hundreds of PRs a day into the Grok @Bot codebase, pstack has allowed me to keep the quality of the code high for everyone as I constantly monitor code, refactor, add new lints and checks, and also work on features.
Being Grok @Bot's gardener and maintainer is something I was only able to do through pstack. Our early momentum after building the prototype was very high and many people were joining the team. I had a critical moment of opportunity to refactor the whole codebase, while it was being built and extended and with no downtime, into something with strong foundations. A codebase with high quality that scales no matter how many engineers (and most importantly, non-engineers) contribute to it. All of this work requires me to refactor and improve the foundations of Grok Bot as it's being built, and you can only do that when the foundations can keep up with the number of contributions.
The proof is in Grok @Bot itself. Over the next few weeks, I'll tell you everything you need to know to be able to build and maintain a high quality app using pstack.
Part 1 β Verification is all you need
The most critical skill to have in your toolbox is a high quality verification skill. This skill is so important to have and maintain that I think of it more like critical infrastructure rather than "just" a skill. A good one will amplify the output of your whole team, including non-engineers. Done well, you will 100-1000x your whole team's output.
If you're not familiar with the term, verification means that an agent can verify its own work. It can keep going until it succeeds at its task, because it can now close the loop without you being the bottleneck. If you're interested to know more of the story of how I created my first verification skill for Cursor, check out my previous post Loops You Can Trust.
Let's build a verification skill together
To start, install pstack and then run /create-verification-skill. I also recommend adding Dr Eggbot, my bot that helps you create high quality bots, to your roster. Dr Eggbot ships with pstack. Itβll teach coding bots how to use it, and it can also make non-coding bots with the same rigor.
You can ask Dr Eggbot to create an engineer bot for you that you can then ask to run /create-verification-skill and set up a daily routine to run /maintain-verification-skill.
While that runs, let's walk through what the skill does and how it makes a high quality verification skill for you.
I distilled all of our verification skills that we use to build Grok @Bot and Cursor into this skill as a sort of meta-skill. It teaches your agent how to create a high quality one for your own app.
Now this is where choice of tech stack is important. If you're building an app in Electron or for the web for example, you can take advantage of the rich debugging tools available for the JS ecosystem. For example, the Chrome DevTools Protocol (CDP) allows you to use the same tooling available in your browser's developer tools. Or if you're building an iOS app, making use of the simulator.
You ideally want the ability to interact with your app, debug it, take perf traces, and any other debugging and development tooling that you might typically use if you were developing the app by hand. If you don't have a rich runtime to make use of, you may need to ask your agent to create tools for you (eg using lldb, or a custom package that runs as a sidecar in dev environments), or just make use of what you have available.
I personally feel that agentic verification is so important that I would unironically suggest building your own rich debugging tools, or even choosing a different tech stack, in order to have unfair advantages and extreme productivity in building software. As I mentioned earlier, giving agents the ability to verify their own work unlocks everyone in your organization to be able to contribute and validate that their changes actually work. The harder your tech stack is to debug and control, the more difficult it will be to use agents productively.
Make it Reproducible
In pstack, we have a principle called "Build the Lever". What this means in the context of cβ¦ continue on X β
β₯ 7.1K Β· β² 648 Β· π 1.2MView on X β
