pstack is Lauren Tan's (poteto) skill set that uses one command to pick the right process for each coding task.
You only type /poteto-mode, because that command calls the other skills when they are needed.
What is pstack for?
pstack · by Lauren Tan (poteto)
Picks a process for each coding task and requires evidence that the result is correct, so you can run several agents in parallel.
You type one command with a description of the task. It picks a playbook and runs the skills each step needs.
23
playbooks
34
situational skills
24
principles
1
command to start
Start with /poteto-mode <task>
How does pstack work?
Take a guess first
In the Bug fix playbook, when can a fix ship?
Every shipped line traces to runtime evidence. A fix without evidence is only a hypothesis.
What do you need to do?
Explored 1/6
The Feature playbook
The agent names the data shape first and verifies the result on the product itself.
howUnderstand the affected subsystemarchitectExplore designs in paralleldelegateA subagent writes the codeverifyVerify on the matching surfaceinterrogateSeveral models look for flaws in a contested design
The Bug fix playbook
Every shipped line traces to runtime evidence.
reproduceThe agent reproduces it itselfhow + whyBinary-search the causefixThe smallest fix the evidence justifiesverifyThe original repro now passestddThe failing repro lands before the fix
The Perf playbook
The agent must tie every fix to a measurement.
baselineCapture a baseline tracebenchmark-checklistVet whether the number can be trustedhowGround the hypothesesmantrasTries the cheapest option first, which is to drop work nothing uses
The Investigation playbook
Read-only. The result is a cited answer.
howHow it workswhyWhy it was built this wayunslopStrips AI tells from the answer
The Autonomous run playbook
You define a checkable exit condition, and the agent runs until it is met.
predicateA checkable exit condition/loopRestarts the work on a schedule or eventshow-me-your-workLogs every iteration
Call /interrogate directly
No playbook needed. Any skill can be called on its own.
interrogateDifferent models try to break it, keeping only findings a run proves
/setup-pstack sets which model plays each role.
How is pstack different from other skill sets?
vs Agent SkillsAgent Skills is organized by project stage. pstack is organized by the kind of task.vs SuperpowersSuperpowers has one path. pstack picks from 23 and cross-checks with several models.vs mattpocock/skillsWith Matt's set you call each skill yourself. With pstack, one command calls the skills for you.
When should you use pstack?
Use it when
Hard work where quality matters more than speed
You want to run many agents in parallel
You use Cursor, or Claude Code with a few steps skipped
Skip it when
Small work that needs little proof
You need low costs, because many subagents and models use a lot of tokens
How do you start with pstack?
bash
/plugin marketplace add cursor/plugins
/plugin install pstack@cursor-pluginsChoose pstack for hard work where proof matters more than speed. Run /setup-pstack once, then start your next task with /poteto-mode. On Claude Code, commands take a pstack: prefix, and pstack skips steps that need a Cursor-only feature.