Coding workflow guide · Cursor & Claude Code
pstack for Cursor: install, setup & FAQ
pstack gives Cursor a structured way to investigate, implement, and verify code changes. This guide covers installation, model setup, and practical prompts for Cursor and Claude Code.
Install the plugin, configure its models with /setup-pstack, and start a substantial task with /poteto-mode. The goal: less unnecessary code and clear evidence that the result works.
Reviewed . Commands and defaults can change. The official pstack README in cursor/plugins is the source of truth for the Cursor plugin.
What is pstack?
pstack is poteto’s engineering skill collection. In the project README, poteto describes experience at Cursor and membership on the React core team, including work on React Compiler.
The project emphasizes quality over raw output. It aims to make Cursor behave more like a disciplined engineering team: understand the problem, choose an approach, delegate deliberately, and verify the result. See the author’s introduction.
There are two repositories to distinguish:
- Cursor plugin: cursor/plugins/pstack.
- Standalone mirror: backnotprop/pstack, distributed under the MIT license and adapted for use across multiple coding agents.
pstack is a workflow and skill collection. Your coding agent and selected models still perform the work.
How is pstack different from casually asking AI to code?
“Fix the settings page” leaves many decisions unstated.
Which behavior is broken? What should remain unchanged? How will anyone know the fix worked?
A more rigorous request gives the agent a target and a way to check it. pstack’s /poteto-mode adds a playbook-driven workflow that can call other skills as needed.
For example, a bug-fix task should produce more than a plausible patch:
| Stage | Useful evidence |
|---|---|
| Understand | Relevant code paths and a clear description of the failure |
| Reproduce | A repeatable example of the incorrect behavior |
| Implement | A focused change tied to the cause |
| Verify | A check demonstrating the intended behavior |
| Explain | What changed, what was checked, and any remaining limits |
That evidence is what makes the work reviewable. Installing a skill alone cannot guarantee it.
How to install pstack
Choose the installation route for the agent you use.
Install pstack in Cursor
Run this in Cursor’s agent chat:
Cursor agent chat
/add-plugin pstackThen configure it:
Cursor agent chat
/setup-pstackStart a task with:
Cursor agent chat
/poteto-modeThese are agent-chat commands, not shell commands. Follow the official Cursor pstack README if the installation flow changes.
Install pstack in Claude Code
Run the following commands inside Claude Code:
Claude Code
/plugin marketplace add cursor/plugins
/plugin install pstack@cursor-pluginsPlugin skills commonly use the pstack: namespace:
Claude Code
/pstack:setup-pstack
/pstack:poteto-mode
/pstack:poteto-helpUse the command names exposed by your installed version. Cursor-specific features and configuration paths may differ in Claude Code. The standalone mirror README documents cross-agent usage and the Claude Code command mapping.
Install through the skills CLI
For a skills-based installation across supported agents, run this in your terminal:
Terminal
npx skills add backnotprop/pstackChoose the skills and target agents offered by the installer.
The mirror documents dependencies between skills. In particular, poteto-mode coordinates other skills, so installing only that single file may leave its workflow incomplete. Review the mirror’s installation and dependency notes when selecting a subset.
How to configure /setup-pstack
In Cursor, /setup-pstack configures model assignments and a reasoning budget. It writes the resulting rule to:
Cursor configuration file
~/.cursor/rules/pstack-models.mdcThe setup documentation describes detecting available models, choosing a budget, reviewing role assignments, and writing the configuration.
Assign models by role
Think about two broad categories:
- Code delegation: implementation work with a sufficiently clear specification.
- Judgment and prose: evaluating tradeoffs, reviewing approaches, and explaining decisions.
Avoid copying an old model list without checking access. Choose models available in your own environment.
Choose a reasoning budget
The setup documentation currently offers these budget labels:
| Budget | Target reasoning level |
|---|---|
small | Medium |
medium | High |
large | Extra high |
unlimited | Maximum |
These labels configure reasoning effort. They do not mean unlimited usage, a fixed price, or guaranteed completion.
Re-run /setup-pstack when your model access or preferences change. For current options and behavior, use the official setup file.
How to use /poteto-mode
Use /poteto-mode at the start of a task that needs deliberate engineering work. It selects a playbook and brings in other skills as the task requires. The poteto-mode documentation defines the workflow.
Give it enough information to make good decisions.
Include five elements in a good prompt
| Element | What to provide |
|---|---|
| Goal | The behavior you want to change |
| Done check | The condition that means the task is complete |
| Proof | The evidence you want returned |
| What you know | Reproduction steps, relevant files, or prior findings |
| Constraints | Scope, compatibility, dependencies, and deployment limits |
Here is an example you can adapt:
Example prompt · Cursor agent chat
/poteto-mode
Goal:
Fix the settings page so a saved timezone survives a page reload.
Done check:
After changing the timezone, saving, and reloading, the selected
timezone and displayed timestamps remain consistent.
Proof:
Show the reproduction before the fix, the verification afterward,
and the relevant test output.
What you know:
The selection updates immediately, but resets after a hard reload.
Start with the settings form and the existing preferences API.
Constraints:
Reuse the current storage mechanism. Add no dependencies.
Keep the API response shape unchanged.
Do not deploy or merge.The prompt defines the outcome and the evidence without prescribing an unverified implementation.
Use /poteto-help when you get stuck
If you are unsure which skill fits, describe the situation:
Example prompt · Cursor agent chat
/poteto-help
The bug disappears when logging is enabled.
Help me choose a workflow to reproduce and investigate it.Consult the poteto-help documentation in the standalone mirror and use the command name supported by your installation.
Common pstack skills
Start with poteto-mode. Learn individual skills when you have a specific reason to invoke them.
| Skill | Useful starting point | Documentation |
|---|---|---|
setup-pstack | Configure models and reasoning budget | Setup |
poteto-mode | Coordinate a substantial engineering task | Mode |
poteto-help | Find guidance for an unfamiliar situation | Help |
how | Understand how an existing subsystem works | How |
architect | Work through an implementation’s structure | Architect |
interrogate | Examine a change through a review workflow | Interrogate |
Skill availability and details vary by version. Browse the official skills directory for the current collection.
pstack FAQ
What does pstack do in Cursor?
pstack adds engineering workflows and skills to Cursor. Its main entry point, /poteto-mode, uses playbooks to coordinate the work. You still need to supply a clear task, appropriate access, and a meaningful completion check.
Who created pstack?
pstack comes from poteto, whose project introduction describes React core and Cursor engineering work. The project’s stated priority is higher-quality code with less unnecessary output. Read the author’s introduction in the official README.
How do I install pstack in Cursor?
Run /add-plugin pstack in Cursor’s agent chat, then run /setup-pstack. Start a substantial task with /poteto-mode. If a command is unavailable, check the current plugin installation and the official README.
What is poteto-mode?
poteto-mode is pstack’s coordinating skill. It matches a task to a playbook and invokes other skills when needed. Use it with a specific goal and observable completion criteria.
Do I need to run every pstack skill manually?
No. The normal starting point is /poteto-mode, which coordinates other skills. Direct commands are useful when you already know you want a particular explanation, configuration step, or review.
Where does pstack store Cursor model settings?
Cursor setup writes model choices to ~/.cursor/rules/pstack-models.mdc. That path is specific to the Cursor setup. Consult the documentation for your agent before editing configuration files.
Can I use pstack with Claude Code?
Yes. Add the cursor/plugins marketplace and install pstack@cursor-plugins. Commands commonly appear with a namespace, such as /pstack:poteto-mode. Check your installed command list and current repository documentation.
Can I install pstack without Cursor?
Yes. The standalone mirror supports skills-based installation through npx skills add backnotprop/pstack. Select the target agents offered by the installer and include dependencies required by the skills you choose.
Is pstack free?
The standalone mirror is MIT-licensed. That license is separate from the cost of Cursor, Claude Code, model access, or usage. Check the relevant providers for current charges.
Does pstack guarantee better code or lower costs?
No. Evaluate it on your own tasks. Compare correctness, review effort, completion time, and actual usage costs. A convincing explanation is useful, but it is not a substitute for checking the resulting behavior.
pstack vs ponytail and other coding skills
pstack vs ponytail
Both projects emphasize avoiding unnecessary code. Ponytail presents a minimal-solution philosophy: reuse what exists and write only what the task needs. pstack emphasizes coordinated engineering playbooks and model roles. A practical comparison is to give each the same bounded task and assess the patch, verification, and review burden. Neither project’s philosophy establishes a universal winner.
pstack vs Claude Code skills
“pstack vs Claude Code skills” compares a particular skill collection with a broader way of extending an agent. pstack can itself be used within Claude Code. The useful choice is between adopting its coordinated workflow and assembling a smaller set of skills for your own process. Start with the behavior you need, then check compatibility, overlapping instructions, and how much configuration you want to maintain.
Related reading on AI Model FAQ
For model selection context, read our Kimi K3.1 guide to coding and agent capabilities. It covers the model rather than pstack integration; confirm compatibility in your own setup.
Related topics planned for future coverage:
- Cursor agent workflows
- Claude Code plugin installation
- Verifying AI-generated code
Get started with pstack
Open the official pstack README, install it in Cursor, and try one small task with a clear done check:
Cursor agent chat
/add-plugin pstack
/setup-pstack
/poteto-modeFor a cross-agent installation, review backnotprop/pstack on GitHub.
AI Model FAQ is an independent publication and is not affiliated with pstack, poteto, Cursor, or Anthropic.