pstack

Installing it

/add-plugin pstack

Read from the plugin's own README, under "install".

The README's own two-step start is /setup-pstack (choose which model runs which role) then /poteto-mode for any task needing rigor. /setup-pstack writes ~/.cursor/rules/pstack-models.mdc; the skill describes it as an override layer over the inline defaults, not a prerequisite.

Cursor’s documented route works for any marketplace plugin regardless: open Customize in the sidebar, find the plugin, select Install, and choose project or user scope.

What it needs

• Cursor, with plugin and Task subagent support — the mode skill uses Cursor's own mode: true and disable-model-invocation frontmatter, so it does not port by copying SKILL.md files.
• Optional: the cursor-team-kit plugin — poteto-mode routes to its /deslop, control-cli, and control-ui skills at specific triggers.
• Optional: Graphite (gt) for the stack-shipping playbooks.

The 44 skills it ships

Each name below is the skill's own directory name, and each description is drawn from that skill's own frontmatter or the plugin's README — not rewritten from the marketplace listing.

• poteto-modemanual onlyThe entry point and the reason the rest exists: a sticky mode that picks a playbook, copies its steps in, and calls the other skills as those steps fire.
• setup-pstackDetects which models your session has available and writes an always-applied rule that sets a model per role, overriding the skill defaults.
• architectmanual onlySketch types, signatures, and module structure before code, then stay in the loop while implementation fills it in.
• arenamanual onlySpawn N parallel candidates at the same task, pick a base, then graft the strongest parts of the losers into it.
• swarmmanual onlyFan out N parallel workers, drain them, and return one report — for coverage, races, gauntlets, and exploration.
• interrogatemanual onlyAdversarial multi-model review: several LLM reviewers challenge a change from independent angles.
• howExplains how a subsystem works, its runtime flow, and where a change belongs — placement, ownership, and layering questions.
• whyDesign rationale and regression archaeology: discovers whichever MCP servers are available at run time and queries each evidence category in parallel.
• blast-radiusmanual onlyFind what a change could break beyond its own diff, and prove the one fact it is safe because of by running real code.
• figure-it-outmanual onlyBuilds an auditable playbook when no narrower one fits — a large migration or an ambitious multi-part change — and scales rigor to the task.
• tddmanual onlyExplicitly narrow: only for a requested failing or regression test, or a bug with an obvious cheap local test target.
• create-verification-skillmanual onlyGenerates a project-local verification skill that drives your app the way a user does, in whatever language or platform it uses.
• maintain-verification-skillmanual onlyA periodic pass that keeps that verification skill and its feature map honest, capped at one PR of proven corrections.
• show-me-your-workmanual onlyKeeps a reviewable decision trail for long or unattended runs — a TSV log, one row per decision, local by default and committed when a reviewer needs it.
• reflectmanual onlySpawns three parallel review subagents over the active transcript and routes each learning to a concrete edit on an existing skill.
• recallmanual onlyReconstructs your recent working context from chat history, live state, and the shared record, and hands back a current-state brief.
• no-commentsmanual onlySpawns Comment Sicko, applies the accepted findings, and offers encodings for any constraint the comments were claiming.
• unslopCuts AI tells from any writing. Its frontmatter says it must always apply.
• technical-writingmanual onlyA layered writing standard — Diátaxis structure, Google developer style, simplified-technical-English instruction rules, Global English syntax.
• teachmanual onlyRuns how and why and weaves both into one plain explanation aimed at a person actually understanding the work.
• bromanual onlyRestates the last message in plain human language, with no jargon.
• automate-memanual onlyDrafts or revises a personal -mode skill capturing how you work, optionally pulling evidence from recent transcripts.
• typescript-best-practicesTypeScript guidance, scoped by its own description to any .ts or .tsx file being read or edited.
• principle-laziness-protocolmanual onlyBias toward deletion and the smallest change that solves the problem.
• principle-foundational-thinkingmanual onlyGet the core data structures right first so downstream code becomes obvious.
• principle-model-the-domainmanual onlyEncode the domain in a structure instead of scattered conditionals.
• principle-fix-root-causesmanual onlyReproduce first, ask why until you reach the root cause, and resist guards that silence crashes.
• principle-prove-it-worksmanual onlyVerify against the real artifact before declaring done — not a proxy, a self-report, or "it compiles."
• principle-guard-the-context-windowmanual onlyRoute bulk work to subagents; keep summaries in the main thread, not raw payloads.
• principle-type-system-disciplinemanual onlyMake illegal states unrepresentable, parse external data at boundaries, and refuse to lie to the compiler.
• principle-boundary-disciplinemanual onlyConcentrate guards at system boundaries and keep business logic in pure functions.
• principle-build-the-levermanual onlyBuild the tool that does the work or proves it, rather than doing it by hand — the tool is what a reviewer can rerun.
• principle-encode-lessons-in-structuremanual onlyWhen you write the same instruction twice, encode it as a lint, flag, runtime check, or script instead of more text.
• principle-exhaust-the-design-spacemanual onlyFor a decision with no precedent, build two or three competing prototypes and compare before committing.
• principle-experience-firstmanual onlyChoose user delight over implementation convenience; ship fewer polished features over more rough ones.
• principle-make-operations-idempotentmanual onlyConverge to the same end state regardless of crashes, restarts, or partial prior runs.
• principle-migrate-callers-then-delete-legacy-apismanual onlyMigrate callers and delete the old API in the same wave instead of keeping compatibility layers.
• principle-minimize-reader-loadmanual onlyCount the layers between question and answer; collapse one-caller wrappers and shrink mutable scope.
• principle-never-block-on-the-humanmanual onlyOn reversible work, proceed and let the human course-correct; reserve confirmation for irreversible actions.
• principle-outcome-oriented-executionmanual onlyIn a planned migration, converge on the target architecture rather than preserving smooth intermediate states.
• principle-redesign-from-first-principlesmanual onlyRedesign as if a new requirement had been foundational from day one, instead of bolting it on.
• principle-separate-before-serializing-shared-statemanual onlyEliminate sharing between concurrent actors first; serialize only when one shared writer is a real invariant.
• principle-sequence-verifiable-unitsmanual onlyBreak work into small units that each end in a verifiable state, and order delivery so the sequence proves itself.
• principle-subtract-before-you-addmanual onlyRemove dead weight and redundant validators first, then build on the simpler base.

Subagents

Cursor subagents the plugin ships. A plugin can carry these; a standalone skill cannot.

• poteto-agentThe routing target for /poteto-mode and for any request asking for this style. Reads the poteto-mode SKILL.md in full, including its inline principles index, before doing any work; resumed rather than duplicated within a conversation.
• comment-sickoSpawned by /no-comments to strip code comments before review — the plugin's own frontmatter calls it "a deranged comment-hater that savors deletion and condemns workaround code."

The 23 playbooks

Named procedures shipped inside a skill rather than as separate plugin components. The mode matches your request to one of these and copies its steps in.

• investigationA read-only question — how X works, why Y was built this way, whether we are sure.
• bug-fixReproduce a defect, root-cause it, and fix it with runtime evidence.
• perf-issueTrace a measured slowness and improve it against a baseline.
• hillclimbSustained improvement of one metric against a target, looping hypotheses with before/after measurement and one commit per accepted win.
• runtime-forensicsDiagnose a live symptom — leak, idle-CPU spin, glitch — from instrumentation.
• trace-forensicsDiagnose a captured profiling artifact: cpuprofile, trace, spindump, heap snapshot.
• featureNew or changed behavior, built from a named data shape.
• refactoringA behavior-preserving change to structure or shape.
• prototypeA throwaway sketch to settle a design or empirical question cheaply by observing it.
• visual-parityPixel-exact UI equivalence between two implementations.
• authoring-a-skillWriting or editing a SKILL.md.
• evalTest how a skill or prompt change affects agent behavior, blinded.
• babysitDrive a PR or a stack to merge-ready: conflicts, review threads, CI.
• shippingIndependently verify a green stack, then land the contiguous verified run with Graphite merge-when-ready.
• autonomous-runDrive a long task to completion without stopping.
• orchestrateA standing project handed to one coordinator chat: multi-day, many stacked PRs, fleets of subagents.
• autopilot-fullRun independent PRs to merged with one owner per PR and root verification of each merge-ready head.
• autopilot-stackBuild and verify one linear Graphite stack for the operator to review and land.
• session-pickupResume or take over a prior agent's in-flight work.
• pause-safelySuspend in-flight work cleanly so it can be resumed later.
• multi-phase-planWork that spans phases or stacked PRs.
• worktree-cleanupReclaim disk by pruning merged or abandoned worktrees and stale iOS simulators, safety-gated.
• opening-a-prShips as a playbook file but is absent from the README's own table of 22.

The caveat that matters

The mode is sticky. Once entered it stays on across turns, applying itself whenever a playbook matches, until you opt out — and its own reminder tells it to stay out of the way on casual turns. Adopting pstack means adopting the process, not adding a tool.

Related entities

Cursor plugins

• Cursor Team KitCursor's own 18-skill kit for CI, code review, shipping, and verification, plus two subagents and two TypeScript rules — deliberately built to work without third-party service integrations.
• ThermosRuns two Cursor subagents in parallel over a branch diff — a correctness/security audit and a strict maintainability rubric — then synthesizes both into one deduplicated, prioritized review.

Skills profiled on this site

• orchestrateFans a large task out across parallel Cursor cloud agents via the Cursor SDK — planners publish tasks, workers hand off, a script reconciles the tree from disk and git.
• create-pluginMeta-workflow for scaffolding a marketplace-ready Cursor plugin — manifest, skills, rules, agents, and commands together — plus a pre-submission quality check.

Sources and provenance

Everything on this page was read from primary sources published by Lauren Tan: the plugin’s own .cursor-plugin/plugin.json manifest, its README, and the frontmatter of every skill, subagent it ships, plus the repository’s marketplace manifest for the quoted listing copy. Last checked on August 20, 2026.

Cursor’s marketplace copy for this plugin reads: “if you want to go fast, go deep first. pstack helps you write less, but higher quality code. rigorous agent workflows you can parallelize with confidence.” — quoted, not assessed.

• The only plugin in cursor/plugins whose manifest names an author other than Cursor: author.name is "Lauren Tan". The repository is the distribution channel, not the author.
• Its README describes 22 playbooks and its table lists 22, but 23 playbook files ship in skills/poteto-mode/playbooks/ — the table omits opening-a-pr. Both figures are the source's own; this page enumerates the 23 files on disk.
• The three skills under automations/benny/ are not counted in the 44. They sit outside the manifest's declared skills directory, so Cursor does not load them as plugin skills — the README presents Benny as a separate issue-triage automation.

Nobody here has installed or run this plugin. Version numbers and component lists are a snapshot and will drift — the source directory is authoritative. Cursor states that every marketplace plugin is manually reviewed before listing; the criteria are not published in detail and have not been audited here.