pstack Skills: Lauren Tan's agent workflow for DSH and Claude Code
pstack is the skill set Lauren Tan (poteto) publishes in cursor/plugins: /poteto-mode with 23 playbooks, 34 skills, 24 principles — and the DSH ports that carry it.
Last updated: 2026-11-16
Quick answer
pstack is the agent skill set that Lauren Tan (@poteto) publishes in the official cursor/plugins repository, under pstack/. Its README states the goal in one line: "the goal is not to maximize loc, in fact it's the opposite. pstack helps you write less, but higher quality code."
Three numbers describe the package, all read from the README and its files on 2026-11-16: 23 playbooks behind /poteto-mode, 34 skills in the "all skills" table (from /poteto-mode down to /technical-writing), and 24 principles in five groups — core, architecture, verification, delegation, meta.
One skill is the front door: /poteto-mode reads your request, picks a playbook, opens a todo list copied from that playbook's steps, and calls the other skills as the steps fire. Everything else is a direct line for when you already know what you want. License: MIT, "Copyright (c) 2026 Lauren Tan".
Who wrote it, and what "official" means here
"i'm poteto. i'm not a president or ceo, but i've worked with millions of lines of code at Meta, Netflix, and Cursor. i'm also on the react core team where i help build and maintain react compiler."
Because the repository is cursor/plugins, pstack is the author's own package hosted by her employer's plugin repository, not a re-upload — and the install lines below are that README's, verbatim. The pitch for parallel work is one sentence there: "pstack gives you fearless parallelism" — if one agent can be trusted to write verifiable code, running several stops being a gamble.
/poteto-mode and its twenty-three playbooks
"it comes with twenty-three playbooks", and the playbook is what decides how much ceremony a task gets. Grouped by what they are for:
| Family | Playbooks |
|---|---|
| Read-only answers | investigation |
| Diagnosis and fix | bug fix, perf, hillclimb, runtime forensics, trace forensics |
| Building | feature, refactoring, prototype, visual parity |
| Meta | authoring a skill, eval |
| Ship a branch | babysit, shipping, opening a pr ("invoked at the end of every other playbook") |
| Long unattended | autonomous run, orchestrate, autopilot-full, autopilot-stack |
| Housekeeping | session pickup, pause safely, multi-phase plan, worktree cleanup |
/poteto-mode also ships two subagents: subagent_type: "poteto-agent", the wrapper every delegate during a run goes through (it reads the mode skill in full first), and comment-sicko, a read-only comment reviewer usually reached through /no-comments.
The 34 skills
The README keeps a table named "all skills", one row per skill. Roughly by job:
- Entry and framing —
/poteto-mode,/poteto-help(finds out what you are trying to do),/in-your-own-words(plays your goal back before it starts),/sharpen(makes the prompt deeper, not bigger). - Understanding a codebase —
/how,/why(answers from commits and PR discussion, not from the code),/recall,/ketchup,/ketchup-video,/teach,/blast-radius. - Writing code —
/architect(settle the caller's usage and types first),/tdd,/figure-it-out(designs a playbook when none fits),/typescript-best-practices. - Parallel and adversarial work —
/arena(N attempts at the same thing),/swarm(N workers across slices),/interrogate(different models try to break your diff). - Proving it works —
/create-verification-skill,/maintain-verification-skill,/create-formal-verification-skill,/maintain-formal-verification-skill,/benchmark-checklist,/statechart,/show-me-your-work. - Writing and cleanup —
/no-comments,/unslop,/bro,/technical-writing.
A note on the name: pstack's /unslop is its own cleanup playbook. The separate unslop skill (heydamir/unslop) is a different project — a UI design skill that de-slops a vibecoded screen — so do not expect one to do the other's job.
- Configuring the package —
/setup-pstack,/automate-me(drafts your own<name>-modeskill),/reflect,/correct,/make-bot-ui.
The two dedicated verification skill pairs (create and maintain, plain and formal) are the least usual part of the list: they generate a project-local verify skill instead of shipping a generic test runner, so the proof outlives the session.
The 24 principles
"twenty-four short skills, one principle each" — /poteto-mode indexes them inline and reads that index at task start. A sample of the rules, because the wording is the point:
| Principle | Group | Rule |
|---|---|---|
laziness-protocol | core | "Bias toward deletion and the smallest change that solves the problem." |
foundational-thinking | core | Get the core types and data structures right so downstream code becomes obvious. |
prove-it-works | verification | Verify against the real artifact — run the feature, read the actual value — not a proxy, self-report or "it compiles." |
fix-root-causes | verification | "Trace each symptom back to the first place the system went wrong and fix it there." |
guard-the-context-window | delegation | "Route bulk to subagents; keep summaries in the main thread, not raw payloads." |
encode-lessons-in-structure | meta | Encode the rule as a lint, metadata flag, runtime check or script instead of more text. |
Install it in Cursor
The README gives exactly one line:
/add-plugin pstack
Then two steps: run /setup-pstack to pick a reasoning budget and the models pstack uses per role, and reach for /poteto-mode whenever a task needs rigor. /poteto-help is the way out when you cannot remember which skill or playbook fits. One Cursor-only habit worth copying: choosing /poteto-mode from the / menu and pressing option+enter (alt+enter on Windows) makes it a custom mode that stays attached every turn, instead of attaching to a single message.
Install it in Claude Code
Claude Code is the second harness the README documents, with two commands:
/plugin marketplace add cursor/plugins
/plugin install pstack@cursor-plugins
Three things change in this harness, all stated in the README:
- Command names gain a prefix. Claude Code prefixes plugin commands and agents, so
/poteto-modeis/pstack:poteto-mode. A dedicated adapter document,pstack/docs/harness/claude-code.md, maps every other command name. - Model choices land in a file.
/pstack:setup-pstackwrites your model selections to~/.claude/rules/pstack-models.mdinstead of Cursor's settings. - Cursor-only steps are skipped. A step that needs a Cursor-only feature — bugbot, automations — is skipped with a one-line note rather than failing the run.
Updates are manual by default: "claude code leaves auto-update off for third-party marketplaces". Turn it on in the /plugin marketplaces tab for cursor-plugins, or update once with claude plugin update pstack@cursor-plugins.
Install it in DSH: the community ports
Here the trail ends and interpretation starts. The pstack README covers Cursor and Claude Code only; it never mentions DeepSeek Harness. Nothing in the package is published for dsh by its author. What exists is two community ports that this directory lists, each with its own record page:
| Port | Its own description | License | Record |
|---|---|---|---|
l246804/dsh-pstack | "pstack for DeepSeek Harness: rigorous agent workflows ported from cursor/plugins." | MIT | record |
aa2246740/pstack-dsh | "DeepSeek Harness port of official pstack. Playbooks and principles are poteto's; only the harness call layer is rewritten." | not stated as a standard SPDX id | record |
Read the second description literally, because it is the honest one: the playbooks and principles are the author's text, the call layer is somebody else's, so command names and subagent wiring may drift from the README you just read. That is the normal shape of a port, and it is why the two records above matter more than any summary — including this one.
If you want the harness basics first, installing skills and how skill installation works explain where a copied skill folder lives, what dsh plugins are covers the plugin side, and the best dsh skills puts pstack next to the other skill packs people install.
Why there is no planning skill
The README answers this itself, and the answer is worth quoting in full:
"cursor already has a great plan mode which works great with pstack. but personally, i don't believe in planning. the best spec is code. if you do want to make a plan,
/poteto-modecovers it, but it's not a default."
So a search for "pstack plan skill" lands on a deliberate absence. The multi-phase plan playbook exists for work that spans phases or stacked pull requests; a skill that produces a plan document does not.
Where it stops
- Planning is out of scope by design — see above.
/poteto-modeis the substitute, and it starts from code. - Four names it references are not bundled.
/deslop,control-cliandcontrol-uiship in thecursor-team-kitplugin, and/create-skillis a Cursor built-in; the README says to installcursor-team-kitalongside pstack for the full set. The benny automation pack is dormant and Cursor-only, and its files are not registered as slash skills. - Model selection is per harness.
/setup-pstackwrites a rule that pins a model per role; in Claude Code that rule is a file under~/.claude/rules/, so the same choices do not travel between harnesses. - A port is a snapshot. The dsh ports carry someone's transcript of the README, and the upstream skill files keep changing.
FAQ
What is pstack?
A plugin published by Lauren Tan (@poteto) inside the official cursor/plugins repository, in the pstack/ directory. Its README calls it "the same skills i use every day to ship high quality code at Cursor", built around one entry skill, /poteto-mode, which matches your task to one of twenty-three playbooks and routes to its other skills as the steps need them. MIT licensed, copyright Lauren Tan.
How do I install pstack in Claude Code?
In Claude Code you add the repository as a marketplace, then install the plugin from it: /plugin marketplace add cursor/plugins, then /plugin install pstack@cursor-plugins. Claude Code prefixes plugin commands, so /poteto-mode becomes /pstack:poteto-mode. It also leaves auto-update off for third-party marketplaces — enable it in the /plugin marketplaces tab, or update once by hand with claude plugin update pstack@cursor-plugins.
Does pstack work in DeepSeek Harness (dsh)?
Not from the author. The pstack README documents Cursor and Claude Code only, and never mentions DSH. What exists for dsh are community ports listed in this directory — l246804/dsh-pstack and aa2246740/pstack-dsh — which keep the playbooks and principles and rewrite the harness call layer. They are third-party work; read each record before you install one.
Is there a planning skill in pstack?
No, deliberately. The README says: "cursor already has a great plan mode which works great with pstack. but personally, i don't believe in planning. the best spec is code." /poteto-mode still carries a multi-phase plan playbook for work that spans phases or stacked pull requests, but nothing in the package exists to write a plan document for its own sake.
Sources
All of the above was read on 2026-11-16:
- github.com/cursor/plugins —
pstack/README.md: the author's introduction, the goal sentence, "pstack gives you fearless parallelism", the/add-plugin pstackline, the two Claude Code commands plus the prefix,~/.claude/rules/pstack-models.mdand auto-update notes, "it comes with twenty-three playbooks" and the playbook table, the "all skills" table, "twenty-four short skills, one principle each" and the principle table, thepoteto-agentandcomment-sickosubagents, the "not shipped here" list, the "why are there no planning skills" answer, the benny pack, and the MIT license line. - This directory's records for the two dsh ports: l246804/dsh-pstack and aa2246740/pstack-dsh. Descriptions and licenses quoted above are taken from those records, not from the upstream repository.
Related reading: the Impeccable skill, the best DeepSeek Harness skills, installing skills, skill installation, what dsh plugins are.
