open-pstack

Where Open Pstack comes from

Open Pstack brings Lauren Tan pstack plugin to Claude Code and Codex. The README explains that pstack began as the set of skills Lauren used to ship code at Cursor, and that Open Pstack is an unofficial community project whose job is to stay as close to her original work as possible while translating the parts that depend on Cursor. The README links to Lauren original repository and notes that Cursor users should stay there, while Claude Code or Codex users should use this one. Open Pstack 1.1.0 tracks pstack 0.14.2 at a specific Cursor commit, and the two projects keep separate version numbers on purpose: the pstack version marks the upstream content, and the Open Pstack version marks the Claude Code and Codex package built from it. The repository preserves the original README unchanged, a technical reference for every skill, an upstream sync record, a change record, and an attribution record. The README is clear that Open Pstack does not promise instant updates and reviews new changes in order, changing only what Claude Code and Codex require. This careful tracking means a reader can see exactly which upstream commit the port is based on and what adaptations were made for the new host environments.

What pstack does

pstack is a plugin for coding agents, not a new model or a hosted service. The README says it gives the agent engineering rules, step by step workflows for different kinds of work, focused skills, and small local tools. The normal entry point is a mode called poteto-mode, where you give a task in plain language and the system reads it, chooses a fitting workflow, learns how the current system works before changing it, and compares designs when the choice matters. It favors small, simple changes over extra machinery, can ask several models to challenge important decisions, runs the code and checks real behavior instead of stopping at passing tests, and carries the work through review, continuous integration, and a ready to merge pull request when asked. The README stresses that pstack does not ask you to trust an agent on day one. It helps the agent leave evidence you can inspect, and suggests starting with supervised work before letting it run more in parallel after its checks have earned that trust in your own repositories. The skills are there to call directly when poteto-mode does not fit, which keeps the system flexible for experienced users. The promise is that the agent shows its work, so a reviewer can follow the evidence instead of trusting a black box result from the start.

Skills and model setup

The README lists useful skills such as how, why, architect, arena, interrogate, create verification skill, maintain verification skill, babysit and reflect, each meant for a specific moment in a task. Setup checks which models you can run, shows how each one will start, and asks before saving choices. The current default group uses four models, one requested effort per model family, and a limit on how many run in parallel. The README notes that Open Pstack does not quietly replace a failed model with a weaker one. Some workflows use a single model, while skills such as architect, arena and interrogate can run several in parallel, with each model using the subscription and token allowance of its own command line tool. Grok can join a multi model review but cannot act as the main app running pstack. Both Claude Code and Codex read the same skills; only the way they start those skills and models differs, and a table in the README spells out those differences side by side. That shared skill tree is a design goal, so behavior stays consistent across the two coding environments. The design keeps Claude Code and Codex on one shared skill tree, so a team using both tools sees consistent behavior from the same set of instructions.

Installing and contributing

Installation inside Claude Code uses plugin marketplace and plugin install commands followed by a reload, while Codex uses its own marketplace and plugin add commands and turns on subagents in its config so pstack can compare work in parallel. A new Codex task is needed after install so it discovers the skills. The README walks through setup-pstack and poteto-mode for both apps, then describes how the other skills are called, either as native skills in Claude Code or by name in Codex. Contributions that fix Claude Code or Codex behavior or bring over new pstack releases are welcome, and the README asks contributors to search existing issues first and to explain why a change belongs in Open Pstack instead of Lauren original project. Pull requests must keep one shared skill tree for both apps and must pass the repository tests, type checks, plugin validation and static checks. The project builds on Michael Denyer pstack-claude port and includes attributed MIT licensed work from Cursor Team Kit and Superpowers. The README points to a technical reference for every skill, dependency and Claude Code or Codex detail. The repository is published under the MIT license and its most recent commit was on 2026-08-26.

Editorial conclusion

Open Pstack is an unofficial community port of Lauren Tan pstack plugin, built to run inside Claude Code and Codex rather than Cursor. The repository is published under the MIT license and its most recent commit was on 2026-08-26.