pstack is a Cursor plugin created by Lauren Tan (poteto) that turns AI coding agents into a structured engineering workflow with 23 workflow skills, 21 engineering principles, 22 task playbooks, and 2 specialized subagents. The central command /poteto-mode routes a request to the right playbook (bug fix, feature, refactor, investigation, perf), copies it verbatim into a task list, and enforces runtime verification rather than accepting a passing build as proof. Other skills include /how and /why for tracing system behavior and history, /architect and /arena for design exploration, /swarm for parallel coverage work, /interrogate for multi-model code review, and /automate-me and /reflect for personalizing and refining the workflow over time. The piece walks through installation, architecture, playbook categories, verification skill generation, overnight autonomous runs, and common mistakes, plus the author's own take on when pstack is worth the overhead versus a lighter workflow like fstack.
Table of contents
Why pstack existsInstall pstackThe pstack architectureWhat the playbooks do/how , /why , /teach , and /recallDesign with /architect and /arena/swarm is different from /arenaMulti-model review with /interrogateThe 21 principlesVerification is a first-class part of the workflowRun pstack while you sleepWhat pstack does not includeMake pstack yoursHow I would use pstackA practical first workflowCommon mistakesMy takeQuestions this post answers
what is pstack and how does it work with Cursor's poteto-mode command
pstack is a Cursor plugin created by Lauren Tan (poteto) built around 23 workflow skills, 21 engineering principles, 22 task playbooks, 2 specialized subagents, and an automation pack. Its main command, /poteto-mode, acts as a router that matches a request to a playbook, copies it verbatim into a task list, delegates work to suitable models, and requires evidence before reporting success, rather than just generating code. daily.dev helps engineers evaluating agentic workflow plugins like pstack track how tools like this evolve.
what is the difference between /arena and /swarm in pstack
Arena gives the same task to several models and synthesizes one result by picking a base candidate and folding in the strongest ideas from others, useful for competing designs. Swarm instead splits a task into independent slices or race arms, waits for every worker, and returns one combined report, useful when you need coverage across many units rather than competing solutions to one brief. developers choosing between competing-design versus coverage workflows can compare approaches like these on daily.dev.
how does pstack verify that a code change actually works instead of just passing the build
pstack rejects a passing build as sufficient evidence and instead matches verification to what changed: a command-line change runs the real command, a UI change walks the changed flow, a migration replays real input, a performance change compares traces, and a storage change reads the value back. When a repository lacks a reliable verification method, the /create-verification-skill command inspects the repo and writes a project-local verification skill covering launch, health check, driving behavior, evidence capture, and cleanup. engineers building trust in AI-driven code changes can follow verification practices like these on daily.dev.