Lauren Tan’s workshop makes a big promise: give agents the right skills and checks, and you can stop reading most of their code. The tool at the center is pstack, a free bundle of skills her team publishes for Cursor. This pstack review looks at what is in the box, how its verification skills work, and which claims from the talk hold up when you check them against the repo and the docs.
What pstack is
pstack lives in the public cursor/plugins repo on GitHub, which is where this pstack review started. A writeup by Barnaby Robson (Oct 5, 2026) counted 26 skills, 24 principles, 23 playbooks and 2 agents in version 0.15.9. Skills are markdown files that the agent loads on demand. Two of them matter most here: create-verification-skill and maintain-verification-skill, added in commit e42d29f on Jul 12, 2026.
How to create verification skill files with pstack
Per that commit, the create skill generalizes her team’s own browser driver into a template. It writes a verify skill for your app under .cursor/skills/, adds a feature map with one file per feature, and then proves itself once end to end. The maintain skill keeps the map current as the product changes; in an X post on Jul 30, 2026, she suggested running it daily on cloud agents. The commit message says the pair was validated by 4 cloud agents against real repos.
pstack review: our 5-point stress test
Each row of this pstack review pairs a claim with the evidence against it.
| # | Claim (talk or vendor) | The other side | Verdict |
|---|---|---|---|
| 1 | Volume proves the method (“i shipped 3000+ PRs in 5 months”) | PR count shows speed; the checks around the agent are what keep the code good | Holds with checks |
| 2 | A verification skill lets you trust the agent | She wrote on X (Oct 12, 2026) that verification is an approximation and can still be wrong | PARTLY, keep hard checks |
| 3 | Grok 4.6 costs the same per token as Grok 4.5 | True for input ($2.00 per 1M) and output ($6.00 per 1M); cached input is $0.50 vs $0.30 per 1M (docs.x.ai, checked Oct 25, 2026) | MOSTLY TRUE |
| 4 | pstack is free and easy to adopt | Free in a public repo, but built for Cursor; other agents need manual porting | TRUE for Cursor users |
| 5 | Agent friendly architecture makes auto merge safe | Works with a strict codebase and CI bans; bans such as no useEffect are her team’s taste, and brownfield code needs a refactor first | DEPENDS on your codebase |
Where the verification skill for ai agents falls short
The skill only checks what the feature map describes. If the map is stale, the agent can prove the wrong thing with confidence. It also assumes your app can be driven by a tool, which is easy for web apps and harder for hardware, data pipelines or anything behind a login wall you can’t automate. And the hill climbing she recommends costs tokens every loop.
Who should use pstack, and our pstack review verdict
Use it if you already work in Cursor and want a starting template rather than a blank page. Write your own if you use other agents; copy the structure (driver, feature map, eval) rather than the files. Either way, the guide on ai agent verification walks through all nine steps.
Why we test the claims before the tools
I’ve built, scaled, and sold three companies, generated over $85 million in the last seven years in my own companies. The tools that burned us were the ones we adopted on a demo. At WKND we read the commit, the price page and the fine print first.
| Evidence | Source |
|---|---|
| create- and maintain-verification-skill, validated by 4 cloud agents | cursor/plugins commit e42d29f, Jul 12, 2026 |
| 26 skills, 24 principles, 23 playbooks, 2 agents in v0.15.9 | Barnaby Robson, Oct 5, 2026 |
| Grok 4.6 cached input $0.50 per 1M vs $0.30 for Grok 4.5 | docs.x.ai, checked Oct 25, 2026 |
Want a verification skill set up on your own app?
We build one verification skill and feature map for your product’s most fragile flow and wire it into CI, under /services/ai-automation-agency/.
FAQ
Is pstack free?
Yes, it is in a public GitHub repo. Running it uses your Cursor plan and model tokens.
Does pstack work outside Cursor?
Not out of the box. The skills are markdown, so you can port them by hand.
Cursor cloud agents review: are they needed?
No. She started locally and moved to the cloud once results held up.
Sources
- Lauren Tan, Grok Bot team workshop, recorded Aug 12, 2026: https://youtu.be/ONeM6YDmvqA
- cursor/plugins, pstack: https://github.com/cursor/plugins
- Barnaby Robson, pstack writeup, Oct 5, 2026
- Lauren Tan on X, Jul 30 and Oct 12, 2026
- xAI, Grok 4.6 model page and pricing, checked Oct 25, 2026: https://docs.x.ai/developers/models/grok-4.6
Who proves your agent’s work before you merge it?
Book a call and we’ll show you the first check worth automating.