What is pstack? Lauren Tan's /poteto-mode for Cursor, explained

pstack is Lauren Tan's open-source skill pack for Cursor: one entry skill, `/poteto-mode`, that matches your task to one of 23 playbooks and calls the other skills as the steps need them. Installing it with /add-plugin pstack changes nothing on its own. The two things that do change your setup are /setup-pstack, which writes one always-applied rule file, and the extra subagents and review panels the playbooks start, which is where the token cost comes from.

Lauren Tan (poteto on X) says in the README that these are "the same skills i use everyday to ship high quality code at Cursor." The pack lives in Cursor's official cursor/plugins repository under an MIT licence. We read version 0.15.15 on 7 October 2026.

Why people are searching for it this week

We checked Google Trends on 7 October 2026, about 08:55 Asia/Riyadh, worldwide, past 7 days:

Over the past 30 days, interest in pstack peaked on 5 October (index 100) and stayed at 79 on 6 October, compared with values between 18 and 62 from 11 to 18 September. Trends values are relative, not search counts.

On X, posts matching pstack, poteto-mode or poteto mode (reposts excluded) numbered 976 in the seven days to 7 October, rising from 59 posts on 2 October to 254 on 5 October (UTC days). Some of those posts are about the unrelated Unix pstack command, so read the count as a direction, not a size. The questions in those threads keep coming back to three things: how to keep the mode on, why it uses so many tokens, and whether it works outside Cursor. This guide answers those three.

pstack at a glance

We counted 51 skill folders in pstack/skills/ on 7 October: the 27 workflow skills listed in the README plus 24 principle-* folders.

What /poteto-mode actually does

When you start a task with /poteto-mode, the README says it does three things:

1. Matches your request to a playbook and copies that playbook's steps into the to-do list. A step it skips stays in the list as skip: <reason>, so you can see what it chose not to do.
2. Calls the other skills as the steps fire. For example, /how explains how a subsystem works, /why looks for why it was built that way, /interrogate asks different models to try to break a diff, and /tdd writes the failing test first.
3. Writes the reply in a plain style, using /unslop to remove AI-sounding phrasing.

The 23 playbooks cover read-only investigation, bug fixes, performance, features, refactoring, prototypes, visual parity, writing a skill, evals, driving a pull request to merge-ready, shipping, long autonomous runs, multi-day orchestration, picking up another agent's session and cleaning up worktrees. The README's own examples show the intended prompt style: a goal plus a check that can pass or fail.

/poteto-mode this pr has a subtle bug where the scroll drifts every 750ms even when idle. repro first, then fix and verify.

The 24 principle skills are one rule each: "Bias toward deletion and the smallest change that solves the problem", "Verify against the real artifact … not a proxy, self-report, or 'it compiles'", and so on. /poteto-mode keeps an index of them and reads it at the start of each task.

What installing changes, and what it doesn't

According to pstack's own help skill, installing changes nothing until you invoke a skill, and only /setup-pstack is meant to load from your words alone. You can check this in the files: poteto-mode, poteto-help and unslop all set disable-model-invocation: true, so the agent will not pull them in on its own. (We explain that switch in When not to auto-invoke a skill.)

Three things can write files:

• `/setup-pstack` writes ~/.cursor/rules/pstack-models.mdc with alwaysApply: true. It maps each role — code work, judgement and prose, the review panels — to a model and records a reasoning budget. Because it sits in your user folder, it applies to every new chat in every project, not just one repository. Delete a line to return that role to its default; delete the file to undo it completely.
• `/create-verification-skill`, which /setup-pstack offers once at the end, writes a project-local skill at .cursor/skills/verify-<app>/ that teaches agents to drive your app the way a user does. You can say no.
• The benny automation pack is not registered as slash skills. It does nothing unless you point Cursor at its setup file, which then copies it into .cursor/automations/benny/ in the target repository.

How do I keep /poteto-mode on?

This is the most common how-to question in this week's posts. Cursor's documentation and the pstack guide give the same answer:

• Plain Enter on /poteto-mode attaches the skill to one message. It fades as the chat moves on.
• Option+Enter (Mac) or Alt+Enter (Windows) turns it into a Custom Mode. It then stays in context on every turn until you exit the mode. The guide notes that Custom Modes are available in the Agents Window and the CLI.

A Japanese post on 6 October, from a reader who has written a book about pstack, warned that from 0.15.10 the mode no longer stays on by itself, so you must use a Custom Mode or type /poteto-mode each turn. That is a user's report, but it matches the current README.

Why does pstack use so many tokens?

Because rigour is done by extra agents. By default, code work (features, refactors, bug fixes, performance) goes to a Grok model, and the hardest changes, prose and judgement go to Claude Opus 5.5. Panel skills such as /arena, /architect and /interrogate run one subagent per model in the list. The default panel has two entries, so each panel step starts two subagents.

Users noticed. "pstack is actually kind of token-hungry huh," one developer posted on 10 September; Lauren Tan's reply was to "try /setup-pstack and choose more efficient models." The setup guide lists four ways to spend less:

Two more practical points. The model rule applies to new chats, so start a new chat after you change it. And if you wrote the rule before an update changed a default, the old default stays pinned until you delete that line or the file and run /setup-pstack again.

Does pstack work in Claude Code or Codex?

Officially, it is built for Cursor. Its help skill says the skills use the Agent Skills format, so other tools can read them, but /poteto-mode, /how, /why and /teach start Cursor subagents with per-role models, and Custom Modes and /loop are Cursor features, so "those parts may not work there."

There is an unofficial port, michael-denyer/pstack-claude, for Claude Code, Codex, Pi and GitHub Copilot. Its README says it tracks upstream and also carries its own "named policy forks", and that it installs a routing hook on Claude Code, Codex and Copilot. A hook joins every session automatically, which is a bigger change than a skill you call by name. It is a third-party port, not published by Lauren Tan or Cursor, so read it before installing it.

A sensible first run

1. Run /add-plugin pstack, then /setup-pstack. If cost matters, pick medium and set the code roles to auto.
2. Start a new chat.
3. Give it something small and checkable. The setup guide's example: /poteto-mode add a --json flag to this command. text output stays byte-identical. verify both.
4. Watch the to-do list. If a step you expected says skip:, read the reason before accepting the result.
5. If you are unsure which skill fits, type /poteto-help with your question. It is written to answer and hand you a prompt, not to start the work.

Who it suits

pstack suits people who already run several agents in parallel in Cursor, want each one to reproduce, verify and review before it reports, and are willing to pay for extra model calls to get that.

It is more than you need if you mostly make small edits, if you are on a tight usage limit, or if you work outside Cursor and don't want to depend on an unofficial port.

If you are comparing workflow packs, Skill Rank already lists Superpowers and Matt Pocock's skills in the coding category. pstack is not in the catalog yet. Stars and buzz do not prove a pack is safe or right for you; the files above do.