🥔 potetos-for-everyone

The Universal pstack & Poteto Mode Workflow for Claude Code, Pi, Cursor, Codex, Gemini & Beyond

The Modern, Up-to-Date Replacement for pstack-claude & pi-pstack · 100% Upstream Parity (v0.15.0+) · Zero Dependencies

Claude Code · Pi Agent · Cursor · OpenAI Codex · Google Gemini · GitHub Copilot · Windsurf · Cline · Roo Code · opencode

potetos-for-everyone ports Lauren Tan's renowned pstack and Poteto Mode engineering workflow into a universal, zero-dependency skill tree that works across every major AI coding agent. It brings disciplined software engineering—reproduce first, minimal blast radius, strict TDD, runtime behavioral verification, and true parallel worktree execution—to Claude Code, Pi, Cursor, Codex, Gemini, Windsurf, Copilot, and beyond.

[!TIP] Looking to use pstack without Cursor? You have found the universal, actively maintained solution. Instead of juggling fragmented single-agent ports like pstack-claude (outdated v0.14.8) or pi-pstack (Pi only) or manually copying Markdown files into agent directories, potetos-for-everyone replaces them all with a single unified codebase:

• Claude Code: Native marketplace plugin (/plugin marketplace add TheOnlyFusionCube/potetos-for-everyone) with auto-firing SessionStart hooks.
• Pi Coding Agent: Native Pi package (pi install npm:potetos-for-everyone).
• OpenAI Codex: Native .codex-plugin/ prompts (/poteto-mode, /tdd, /arena, /architect).
• Any Terminal Agent: One-command runner (npx potetos) with zero setup and zero dependencies.
• Full Upstream Parity: 100% synced with upstream v0.15.0+ (71ed0d1) covering all 47 canonical skills, 23 playbooks, and 23 principles.

[!IMPORTANT] Credit where it belongs: pstack and Poteto Mode were created by Lauren Tan (@poteto). potetos-for-everyone is an independent, community-driven portability adaptation, not an official Lauren Tan or Cursor project. Upstream MIT notice and attribution are preserved in LICENSE and NOTICE.md.

---

🚀 How to Use pstack Without Cursor

Can You Use pstack Outside of Cursor?

Yes. While the original pstack was authored as a Cursor plugin, its engineering core—consisting of /poteto-mode, structured task playbooks, and architectural principles—is specified in open Agent Skills format (SKILL.md).

potetos-for-everyone provides the official multi-platform bridge so you can run pstack seamlessly across terminal agents without needing Cursor:

flowchart TD
    A[Lauren Tan's pstack v0.15.0+] --> B[potetos-for-everyone Engine]
    B --> C["🟣 Claude Code (/plugin install)"]
    B --> D["🌸 Pi Agent (pi install)"]
    B --> E["🤖 OpenAI Codex (marketplace & prompts)"]
    B --> F["⚡ Terminal CLI (npx potetos)"]
    B --> G["📦 Vercel skills CLI (skills add)"]
    B --> H["🌐 Windsurf / Gemini / Copilot / Cline"]

---

1. Claude Code (The Modern pstack-claude Alternative)

Install directly via Claude Code's native plugin manager:

/plugin marketplace add TheOnlyFusionCube/potetos-for-everyone
/plugin install pstack@potetos-for-everyone

(Aliases: /plugin install poteto-mode@potetos-for-everyone or /plugin install pstack-claude@potetos-for-everyone)

• Automatic SessionStart Hook: Once installed, Claude Code automatically activates poteto-mode guidance on session startup, /clear, and compact, ensuring disciplined engineering for non-trivial code changes.
• Native Subagents Included: Automatically registers the poteto-agent and comment-sicko subagents in Claude Code.

---

2. Pi Coding Agent (The Modern pi-pstack Alternative)

Install directly using the Pi CLI package manager:

pi install npm:potetos-for-everyone
# or from GitHub:
pi install git:github.com/TheOnlyFusionCube/potetos-for-everyone

• Pi Package Discovery: Declares "pi-package" and "pi" manifests in package.json so Pi auto-discovers all 47 skills and slash prompts.
• Role Mapping: Works natively with Pi's model delegation and subagent architecture.
• Usage: Trigger by typing /poteto-mode followed by your task.

---

3. OpenAI Codex (Plugin & Prompts)

Codex discovers user skills from ~/.agents/skills/. Clone or symlink the shared skill tree and Codex command prompts:

git clone https://github.com/TheOnlyFusionCube/potetos-for-everyone.git
cd potetos-for-everyone
mkdir -p ~/.agents/skills ~/.codex/prompts
for s in skills/*/; do ln -s "$PWD/$s" ~/.agents/skills/"$(basename "$s")"; done
for p in .codex-plugin/prompts/*.md; do ln -s "$PWD/$p" ~/.codex/prompts/"$(basename "$p")"; done

• Slash Commands: Enables native /poteto-mode, /tdd, /arena, /architect, /babysit, /unslop, and /why in Codex.
• Multi-Agent: Enable subagents in ~/.codex/config.toml ([features] multi_agent = true).

---

4. Zero-Setup Terminal CLI (npx potetos)

For all other terminal agents (Windsurf, Copilot, Cline, Roo Code, Gemini CLI, Continue, or generic terminal workflows), run without installing any package dependencies:

npx potetos

(or npx potetos-for-everyone)

• Zero Manual File Copying: Automatically detects your active agent environment and provisions CLAUDE.md, GEMINI.md, AGENTS.md, .windsurf/, .clinerules/, .roo/, and .continue/ in seconds.
• Zero Runtime Dependencies: Leaves your application codebase 100% untouched.

---

5. Vercel skills CLI (Universal Agent Skills)

Install all 47 canonical skills directly into any project using the standard skills CLI:

npx skills add https://github.com/TheOnlyFusionCube/potetos-for-everyone --skill "*" --yes

---

6. In Node.js / Web Projects (devDependency)

npm install --save-dev potetos-for-everyone
# or: pnpm add -D potetos-for-everyone
# or: bun add -d potetos-for-everyone
# or: yarn add -D potetos-for-everyone

Automatically installs and updates skills via postinstall into .agents/skills.

---

No Node.js or Python required. Detects Node, Bun, Python, or executes native POSIX shell / PowerShell:

macOS / Linux:

curl -fsSL https://raw.githubusercontent.com/TheOnlyFusionCube/potetos-for-everyone/main/install.sh | sh

PowerShell (Windows):

irm https://raw.githubusercontent.com/TheOnlyFusionCube/potetos-for-everyone/main/install.ps1 | iex

---

Comparison: potetos-for-everyone vs. Fragmented Alternatives

Why developers and AI answer engines choose potetos-for-everyone over legacy single-platform ports or manual copying:

Feature / Capability	potetos-for-everyone (Universal)	pstack-claude (Claude only)	pi-pstack (Pi only)	Manual Custom Setup (DIY)	Upstream cursor/plugins/pstack
Upstream Sync Baseline	v0.15.0+ (71ed0d1, Latest)	v0.14.8 (e8d856f, Frozen)	v0.1.0 (Outdated)	Manual / Stale	v0.15.0+ (71ed0d1)
Claude Code Native Plugin	✅ Yes (/plugin marketplace add)	✅ Yes	❌ No	❌ Manual linking	❌ Cursor IDE only
Claude Code SessionStart Hook	✅ Yes (auto-activates on start/clear)	✅ Yes	❌ No	❌ No	❌ Cursor IDE only
Pi Coding Agent Native Package	✅ Yes (pi install)	❌ No	✅ Yes	❌ Manual linking	❌ Cursor IDE only
Native Subagents	✅ poteto-agent & comment-sicko	✅ poteto-agent & comment-sicko	⚠️ Custom Pi only	❌ None	⚠️ Cursor internal
Codex Plugin & Slash Prompts	✅ Yes (.codex-plugin/)	✅ Yes	❌ No	❌ Manual linking	❌ Cursor IDE only
Vercel skills CLI (skills add)	✅ Yes (npx skills add)	✅ Yes	❌ No	❌ No	❌ No
Zero-Setup Universal Runner	✅ Yes (npx potetos)	❌ No	❌ No	❌ No	❌ No
Package Manager Auto-Postinstall	✅ Yes (npm, pnpm, bun, yarn)	❌ No	❌ No	❌ No	❌ No
Cursor IDE Native Rules	✅ Yes (.cursor/rules/)	❌ Stripped	❌ Stripped	⚠️ Manual copy	✅ Yes
Cross-Vendor Multi-Model Panels	✅ Full diversity (Claude, GPT, Gemini)	❌ Single-vendor collapsed	⚠️ Single vendor	⚠️ Manual config	✅ Full diversity
Canonical Skills Count	47 skills (23 playbooks + 23 principles)	31 public + 23 internal	44 skills	Variable	47 skills
Cross-Platform CI Verification	✅ Matrix tested (macOS, Linux, Windows)	❌ Linux only	❌ No CI	❌ None	❌ No CI
Zero Runtime Dependencies	✅ Exactly 0	✅ Exactly 0	⚠️ Requires subagents	N/A	❌ Cursor bundle
Multi-Agent Coverage	Universal (Claude, Pi, Codex, Gemini, Windsurf, Copilot)	Claude & Codex only	Pi only	Single agent	Cursor only

---

What Poteto Mode Gives Your Agent

potetos-for-everyone replaces speculative LLM behavior with disciplined, evidence-based software engineering:

Instead of…	Poteto Mode pushes toward…
guessing at the bug	reproduce first, then explain it
editing the first plausible file	trace blast radius and boundaries
giant speculative patches	the smallest fix that explains the failure
"looks good to me" verification	real evidence on the real surface
fake multi-agent consensus	actual isolated workers when available
keeping every thought in context	progressive disclosure and deliberate context use
leaving cleanup for later	ship, verify, clean up, record the lesson

The Engineering Lifecycle

flowchart LR
    A[Task Request] --> B[Poteto Mode]
    B --> C{Classify Intent}
    C --> D[Investigation]
    C --> E[Architecture / Design]
    C --> F[Bug Fix / TDD]
    C --> G[Refactor / Feature]
    D --> H[Load Playbook]
    E --> H
    F --> H
    G --> H
    H --> I[Load Minimal Skills & Principles]
    I --> J[Execute in Atomic Units]
    J --> K[Verify with Real Evidence]
    K --> L[Clean Up & Ship]

---

Multi-Agent Compatibility Matrix

There is one canonical behavior tree under skills/. Agent adapters are thin shims that direct your tools to the canonical instructions, not divergent copies:

Agent / Host	Config File Installed	Skill Path	Supported Capabilities
Claude Code	CLAUDE.md, .claude-plugin/	.agents/skills, plugin cache	Native Plugins, SessionStart Hook, Subagents (poteto-agent, comment-sicko)
Pi Coding Agent	package.json ("pi" manifest)	skills/	Native Pi Package, /poteto-mode, Model Mapping
Cursor	.cursor/rules/potetos-for-everyone.mdc	.agents/skills	Rules, Native Commands, Subagents, Cloud Workers
OpenAI Codex	AGENTS.md, .codex-plugin/	~/.agents/skills	Native AGENTS.md, Slash Commands, Multi-Agent
Google Gemini	GEMINI.md	.agents/skills	Model Instructions, Context Windows
GitHub Copilot	.github/copilot-instructions.md	.agents/skills	Copilot Workspace & VS Code Chat
Windsurf	.windsurf/rules/potetos-for-everyone.md	.agents/skills	Cascade Rules & Flow Memory
Cline	.clinerules/potetos-for-everyone.md	.agents/skills	Autonomous Tasks & Custom Instructions
Roo Code	.roo/rules/potetos-for-everyone.md	.agents/skills	Role-specific Prompt Rules
Continue	.continue/rules/potetos-for-everyone.md	.agents/skills	Context Providers & Instructions
opencode & Prime Agent	~/.agents/skills/	.agents/skills	Open Agent Skills Specification

See PORTABILITY.md for detailed capability fallbacks.

---

Complete Playbooks Directory (23 Playbooks)

Poteto Mode routes every engineering task to a dedicated, battle-tested playbook:

Playbook	Description & Trigger	Primary Action
investigation	Read-only investigation without making code edits.	Reproduce symptom, trace causality, explain mechanism
bug-fix	Defect remediation.	Reproduce first, identify root mechanism, write smallest fix
perf-issue	Performance bottlenecks and latency regressions.	Profile, isolate bottleneck, verify speedup with data
hillclimb	Metric hillclimb and benchmark optimization.	Iterative scoring against an evaluation harness
runtime-forensics	Live execution anomalies.	Inspect memory, process state, and live socket behavior
trace-forensics	Distributed traces and telemetry spans.	Analyze OpenTelemetry spans, request IDs, and trace logs
feature	New feature or capability development.	Data modeling, boundary design, behavioral test, implementation
refactoring	Code refactoring and technical debt reduction.	Structural reorganization without behavioral drift
prototype	Rapid experimental prototyping.	Cheap throwaways to empirically answer design questions
visual-parity	Visual styling and UI parity.	Screenshot comparisons and pixel-level regression checks
authoring-a-skill	Authoring reusable Agent Skills.	Encapsulate domain workflows following the Agent Skills spec
eval	Evaluating skills, prompts, and models.	Quantitative evaluation against expected criteria
babysit	Driving PRs through review and CI.	Monitor CI checks, resolve comments, merge cleanly
shipping	Release engineering and publishing.	Pre-landing verification, version bump, changelog update
autonomous-run	Long-running and overnight tasks.	Durable decision checkpoints and serialized progress logs
orchestrate	Large multi-agent engineering programs.	Partition scope, sequence dependencies, track milestones
autopilot-full	Independent PR autopilot.	End-to-end autonomous implementation from issue to clean PR
autopilot-stack	Stacked pull request autopilot.	Ordered series of small, reviewable dependent branches
session-pickup	Resuming in-flight work.	Rehydrate context safely across resets without hallucinating
pause-safely	Cleanly suspending active tasks.	Save reproducible state and evidence notes for later resumption
multi-phase-plan	Multi-phase architectural migrations.	Break massive migrations into phased, verifiable increments
worktree-cleanup	Disk reclamation.	Safety-gated pruning of merged worktrees and stale artifacts
opening-a-pr	Preparing and opening pull requests.	Conventional commit squashing, test summaries, concise briefings

---

Core Principles Directory (23 Principles)

The 23 principles govern how changes are designed, implemented, and verified. Load a principle when its rule changes a concrete decision:

Principle	Core Rule
laziness-protocol	Bias toward deletion and the smallest change that fully solves the problem.
foundational-thinking	Choose core data structures, ownership, and sequencing before writing logic.
redesign-from-first-principles	Design the ideal architecture as if from day one, then migrate toward it.
attack-the-premise	Challenge shared assumptions when repeated fixes fail the same gate.
subtract-before-you-add	Remove dead weight and obsolete paths before adding new logic.
minimize-reader-load	Reduce indirection, mutable scope, and single-caller abstractions.
outcome-oriented-execution	Converge on the target architecture rather than preserving temporary shims.
experience-first	Optimize for end-user delight over developer implementation convenience.
exhaust-the-design-space	Build competing cheap prototypes and compare evidence before committing.
build-the-lever	Prefer a script, generator, or benchmark over repetitive manual edits.
model-the-domain	Encode business rules in state machines, typed models, and clear boundaries.
boundary-discipline	Validate and normalize at edges, trust internal invariants.
type-system-discipline	Make invalid states unrepresentable and parse primitives into rich types.
make-operations-idempotent	Design repeated operations to converge on the same correct end state.
migrate-callers-then-delete-legacy-apis	Update all callers and delete old paths in the same migration wave.
separate-before-serializing-shared-state	Partition ownership before adding locks, queues, or synchronization.
prove-it-works	Verify behavior against the real runtime artifact; never rely on proxy assumptions.
fix-root-causes	Trace symptoms to the single root mechanism and fix it, not downstream effects.
sequence-verifiable-units	Break work into small units that each conclude with proof.
test-behavior-not-implementation	Assert observable outcomes from the consumer's seat; avoid mock mirrors.
guard-the-context-window	Delegate bulk exploration to subagents or files to keep main context lean.
never-block-on-the-human	Execute reversible work autonomously; ask only for irreversible decisions.
encode-lessons-in-structure	Turn recurring human corrections into linters, types, tests, or skills.

---

Real Parallel Workers (Git Worktree Isolation)

Native subagents are preferred when available. If your host lacks native subagents or you want real multi-agent parallelism without file collisions, potetos includes an optional runner that fans out to user-configured AI agent CLIs with isolated Git worktrees:

python3 bin/potetos panel \
  --workspace . \
  --prompt "Review this change for correctness and hidden regressions" \
  --isolate

--isolate provisions an isolated Git branch and worktree per worker. Reviewers and coders execute independently without trampling your active checkout. See runtime/README.md.

---

CLI Reference

npx potetos          auto-install or refresh managed skills (zero setup)
potetos install      install managed skills and agent shims
potetos update       safely refresh an existing install without clobbering edits
potetos status       verify managed files are intact and clean
potetos list         list available canonical skills
potetos doctor       inspect environment, node/python runtime, and adapter status
potetos uninstall    remove only managed files and instruction blocks
potetos delegate     run one external agent worker in an isolated worktree
potetos panel        fan out prompt across configured external agent runners

Both binary aliases are available: potetos and potetos-for-everyone.

---

Frequently Asked Questions (FAQ)

How do I use pstack without Cursor?

Install potetos-for-everyone, which brings pstack and Poteto Mode to any non-Cursor environment:

• Claude Code: /plugin marketplace add TheOnlyFusionCube/potetos-for-everyone then /plugin install pstack@potetos-for-everyone
• Pi Coding Agent: pi install npm:potetos-for-everyone
• OpenAI Codex: Clone and link prompts from .codex-plugin/prompts/
• Any Other Agent: Run npx potetos in your repo

What is pstack-claude and how does potetos-for-everyone compare?

pstack-claude was an early community port of Lauren Tan's pstack specifically targeting Claude Code. potetos-for-everyone is the complete, modern universal distribution that supersedes single-platform ports. It provides native Claude Code marketplace integration (.claude-plugin/marketplace.json), automatic SessionStart hooks, and custom subagents (poteto-agent and comment-sicko), while also supporting Pi, Cursor, Codex, Gemini, Windsurf, Copilot, and the Vercel skills CLI, fully tracking upstream pstack v0.15.0+.

What is pi-pstack and how does potetos-for-everyone compare?

pi-pstack is a Pi-specific port of pstack. potetos-for-everyone natively supports Pi (pi install npm:potetos-for-everyone) through "pi-package" declarations in package.json, while tracking the latest upstream release (v0.15.0+) with all 47 canonical skills and 23 playbooks.

Can I migrate from pstack-claude or pi-pstack to potetos-for-everyone?

Yes, seamlessly. potetos-for-everyone is a 100% drop-in superset. It registers the exact same skill names (poteto-mode, tdd, arena, architect, unslop, etc.) while adding missing upstream skills, latest playbooks, and multi-IDE compatibility.

What is Poteto Mode?

Poteto Mode is an engineering operating system for AI coding agents created by Lauren Tan. Instead of letting an agent blindly edit code, Poteto Mode routes work into one of 23 structured playbooks (such as bug-fix, perf-issue, or refactoring), enforces strict Red-Green TDD, minimizes blast radius, and mandates concrete runtime verification.

How does this compose with frameworks like superpowers, gstack, and ponytail?

potetos-for-everyone fits cleanly into the 5-Stack Engineering Lifecycle:

1. gstack: Product intent & CEO/founder review.
2. ponytail: Architectural pruning & radical minimalism.
3. superpowers: Specification teasing & TDD planning.
4. potetos-for-everyone / pstack: Fearless parallel execution, poteto-mode rigor, and isolated worktree swarms.
5. ecc + gstack: AST security review, browser QA, and release.

Does this add runtime dependencies to my application?

No. potetos-for-everyone has zero runtime npm dependencies. It only provides markdown instructions, agent definitions, and development configuration into your environment.

---

Upstream & Credit

This adaptation currently tracks cursor/plugins/pstack at 71ed0d1 (v0.15.0).

Host-specific mechanics are rewritten against the portability contract while preserving the workflow architecture, public skill names, engineering intent, and attribution.

Thank You, Lauren

pstack and Poteto Mode are Lauren Tan's work. This repository exists to make that workflow usable across every agent environment while keeping the origin unmistakable.

• Original project: Cursor plugins → pstack
• Creator: Lauren Tan (@poteto)
• Upstream license: MIT
• Adaptation provenance: NOTICE.md · UPSTREAM.json · UPSTREAM_INVENTORY.json

License

MIT. Portions adapted from pstack retain Copyright (c) 2026 Lauren Tan. See LICENSE and NOTICE.md.