# pstack for Claude Code

> Lauren Tan's (poteto's) rigorous-agent skill stack, published as a Cursor plugin (v0.15.5, 12d587d), and how to run it under Claude Code. Source: https://github.com/cursor/plugins/tree/main/pstack

How to use this file: pick the skill whose `when` matches the task, fetch its SKILL.md from the raw URL, and follow it. For any non-trivial task start with poteto-mode, which routes to the rest. Apply the adapter below first; the skills name Cursor tools.

## Install

```bash
# 1. Fetch pstack (sparse clone: only the pstack/ directory)
git clone --depth 1 --filter=blob:none --sparse \
  https://github.com/cursor/plugins.git /tmp/cursor-plugins
git -C /tmp/cursor-plugins sparse-checkout set pstack

# 2. Install for every project (use .claude/ in a repo to share with a team)
mkdir -p ~/.claude/skills ~/.claude/agents
cp -R /tmp/cursor-plugins/pstack/skills/* ~/.claude/skills/
cp /tmp/cursor-plugins/pstack/agents/*.md ~/.claude/agents/

# 3. Drop the Cursor-only skills
rm -rf ~/.claude/skills/make-bot-ui ~/.claude/skills/setup-pstack

# 4. Slash command must be kebab-case
sed -i.bak 's/^name: Poteto Mode$/name: poteto-mode/' \
  ~/.claude/skills/poteto-mode/SKILL.md

# 5. Let Claude invoke the skills poteto-mode routes to. Every pstack
#    skill ships disable-model-invocation: true, which in Claude Code
#    means only YOU can run it — poteto-mode could not call /how.
for s in how why teach architect arena swarm interrogate blast-radius \
         tdd typescript-best-practices no-comments unslop \
         technical-writing figure-it-out show-me-your-work; do
  sed -i.bak '/^disable-model-invocation: true$/d' \
    ~/.claude/skills/$s/SKILL.md
done
find ~/.claude/skills -name 'SKILL.md.bak' -delete

# 6. Paste the adapter block below into ~/.claude/CLAUDE.md
```

## Adapter (paste into ~/.claude/CLAUDE.md)

```markdown
## pstack (Cursor skills running in Claude Code)

pstack skills in ~/.claude/skills/ were written for Cursor. Translate:

- "Task tool" → the Agent tool. subagent_type "generalPurpose" → "general-purpose".
- "AskQuestion" → AskUserQuestion. "create-skill" → edit SKILL.md directly.
- ~/.cursor/projects/ → ~/.claude/projects/. .cursor/skills/ → .claude/skills/.
- /deslop → /simplify. control-ui → Playwright MCP. control-cli → Bash.
- Principles (principle-*) are read, not invoked: ~/.claude/skills/<name>/SKILL.md.
- Where a skill says to read ~/.cursor/rules/pstack-models.mdc, use this table.
  A value of "inherit" means omit model. Panels spawn one subagent per entry.

feature, refactoring: sonnet
bug-fix: sonnet
perf-issue: sonnet
hillclimb: sonnet
judgment and prose: opus
hardest tasks: opus
how explorer: sonnet
how explainer: opus
why investigators: sonnet
why synthesizer: opus
reflect tooling: sonnet
reflect judgment, divergent, synthesizer: opus
arena runners: opus, sonnet, fable
arena cross-judge pool: opus, sonnet, fable
swarm workers: sonnet
architect runners: opus, sonnet, fable
interrogate reviewers: opus, sonnet, fable
```

## Skills

- [poteto-mode](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/poteto-mode/SKILL.md) — Entry. When: The default entry point for any non-trivial task. Matches the request to one of 23 playbooks, copies its steps into the todo list, and routes to the other skills as steps need them. Claude Code: adapter — Rename `name: Poteto Mode` to `poteto-mode`. Not sticky in Claude Code (Cursor's `mode: true` is ignored), so re-invoke it or say "stay in poteto-mode". `/deslop` → `/simplify`; `control-ui` → Playwright MCP; `control-cli` → Bash.
- [figure-it-out](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/figure-it-out/SKILL.md) — Entry. When: No bundled playbook fits, or the work is a large migration or something you will review after stepping away. Designs a bespoke, auditable playbook and logs decisions. Claude Code: works as-is
- [how](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/how/SKILL.md) — Understand. When: You want a walkthrough of how a subsystem works, or where something should live and which layer owns it. Claude Code: adapter — Explorer and explainer subagents: model roles via the adapter.
- [why](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/why/SKILL.md) — Understand. When: You want to know why something was built this way. Queries every evidence source your MCPs expose (source control, tracker, docs, chat, observability, errors, analytics) in parallel and cites them. Claude Code: adapter — Uses whatever MCP servers Claude Code has connected; model roles via the adapter.
- [teach](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/teach/SKILL.md) — Understand. When: You want to actually understand a change or subsystem, not a summary. Runs how + why and builds one plain explanation diagram by diagram. Claude Code: works as-is
- [recall](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/recall/SKILL.md) — Understand. When: You are starting or resuming work and want your recent context on a topic rebuilt from your chat history and the shared record. Claude Code: adapter — Chat history lives in `~/.claude/projects/<project>/*.jsonl`, not `~/.cursor/projects/`.
- [architect](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/architect/SKILL.md) — Design. When: You are about to write code that crosses a function boundary and want the caller's usage, types and module shape settled first. Claude Code: adapter — Runner panel via the adapter's `architect runners` line.
- [arena](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/arena/SKILL.md) — Design. When: You want N parallel attempts at the same thing, a judge on a different model, then the best parts grafted onto one base. Claude Code: adapter — Candidates run as Agent-tool subagents, each with `isolation: "worktree"`. Model diversity is Claude-only (Fable / Opus / Sonnet).
- [swarm](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/swarm/SKILL.md) — Design. When: You want N parallel workers across different slices, coverage lanes or race arms, then one aggregated PASS / ISSUES / BLOCKED report. Claude Code: adapter — Workers are Agent-tool subagents; `swarm workers` via the adapter.
- [interrogate](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/interrogate/SKILL.md) — Design. When: You have a diff and want several models to try to break it, including a strict code-quality lens. Findings two models raise independently are the high-signal ones. Claude Code: adapter — Reviewer panel via the adapter. Without Grok/GPT the panel is one model family, so consensus is weaker evidence than in Cursor.
- [blast-radius](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/blast-radius/SKILL.md) — Design. When: A small-looking change you don't trust. Finds what else it could break and proves the one fact it's safe because of by running code. Claude Code: works as-is
- [tdd](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/tdd/SKILL.md) — Build & clean. When: Fixing a bug with a cheap local test path. Failing test first, then the fix. Skips itself when a test would need a heavy harness. Claude Code: works as-is
- [typescript-best-practices](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/typescript-best-practices/SKILL.md) — Build & clean. When: Reading or editing .ts / .tsx. Grounds type-system discipline in syntax: discriminated unions, unknown at boundaries, exhaustive variants. Claude Code: works as-is — Its `paths: ["**/*.ts", "**/*.tsx"]` frontmatter is honored by Claude Code.
- [no-comments](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/no-comments/SKILL.md) — Build & clean. When: Before review. Spawns the Comment Sicko agent, fixes accepted findings, and offers code encodings for constraints comments claimed. Claude Code: works as-is — Needs `agents/comment-sicko.md` in `~/.claude/agents/`.
- [unslop](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/unslop/SKILL.md) — Build & clean. When: Cleaning up any writing: removes AI tells. Applies to every prose surface, including the agent's own reply. Claude Code: works as-is
- [technical-writing](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/technical-writing/SKILL.md) — Build & clean. When: Docs, RFCs, readmes, PR descriptions, commit messages. Layered standard: Diátaxis, Google developer style, STE, Global English. Claude Code: works as-is
- [bro](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/bro/SKILL.md) — Build & clean. When: You want the last message restated in plain human language, no jargon. Claude Code: works as-is
- [create-verification-skill](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/create-verification-skill/SKILL.md) — Verify. When: Your project has no scripted way to prove app behavior. Generates a project-local verify skill (Launch, Doctor, Drive, Evidence, Cleanup) plus a feature map, and proves it once end to end. Claude Code: adapter — Write to `.claude/skills/verify-<app>/`, not `.cursor/skills/`.
- [maintain-verification-skill](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/maintain-verification-skill/SKILL.md) — Verify. When: The verify skill's feature map has drifted from the app. One source reader per feature plus one live pass; ends clean, changed (one PR) or blocked. Claude Code: adapter — Reads `.claude/skills/verify-<app>/`.
- [show-me-your-work](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/show-me-your-work/SKILL.md) — Long runs. When: Long, unattended or multi-phase work. Keeps a TSV decision trail (what, why, evidence, result) you can audit in the morning. Claude Code: works as-is
- [reflect](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/reflect/SKILL.md) — Meta. When: A long task landed and you want the recipe captured. Three parallel reviewers read the transcript and route each learning to a concrete skill edit. Claude Code: adapter — Transcript in `~/.claude/projects/`; skills in `~/.claude/skills/`; `create-skill` → edit SKILL.md directly or use `skill-creator`.
- [automate-me](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/automate-me/SKILL.md) — Meta. When: You want your own `<you>-mode` skill drafted from how you have actually worked, routing through pstack underneath. Claude Code: adapter — Mines `~/.claude/projects/*/*.jsonl`; writes to `~/.claude/skills/`; `AskQuestion` → `AskUserQuestion`.
- [setup-pstack](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/setup-pstack/SKILL.md) — Meta. When: Picks which model each role uses and at what reasoning budget, written as an always-applied Cursor rule. Claude Code: replaced — Writes `~/.cursor/rules/pstack-models.mdc` from Cursor's model list. Use the adapter's role table instead.
- [make-bot-ui](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/make-bot-ui/SKILL.md) — Meta. When: A page or dashboard whose buttons wake a Grok Bot over a webhook, including the sender-key handoff and Tailscale. Claude Code: skip — Specific to Cursor's Grok Bot webhooks.

## Principles

Read, not invoked. poteto-mode names each principle that shaped a decision.

- [principle-laziness-protocol](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-laziness-protocol/SKILL.md) — core. Bias toward deletion and the smallest change that solves the problem.
- [principle-foundational-thinking](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-foundational-thinking/SKILL.md) — core. Before writing logic, get the core types and data structures right so downstream code becomes obvious; sequence scaffold before feature; ask what concurrent actors share.
- [principle-redesign-from-first-principles](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-redesign-from-first-principles/SKILL.md) — core. Redesign as if the new requirement had been a foundational assumption from day one, instead of bolting it on.
- [principle-attack-the-premise](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-attack-the-premise/SKILL.md) — core. When two fixes sharing one premise fail the same gate, take a census of who holds the imbalance and question the premise instead of writing a third fix.
- [principle-subtract-before-you-add](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-subtract-before-you-add/SKILL.md) — core. Remove dead code, redundant validators and stub references first, then build on the simpler base.
- [principle-minimize-reader-load](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-minimize-reader-load/SKILL.md) — core. Count layers between question and answer and hidden state in the reader's head; collapse one-caller wrappers and shrink mutable scope.
- [principle-outcome-oriented-execution](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-outcome-oriented-execution/SKILL.md) — core. In planned rewrites and migrations, converge on the target architecture; don't preserve smooth intermediate states with throwaway compatibility code.
- [principle-experience-first](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-experience-first/SKILL.md) — core. Choose user delight over implementation convenience; ship fewer polished features over more rough ones.
- [principle-exhaust-the-design-space](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-exhaust-the-design-space/SKILL.md) — core. For a novel interaction or architecture with no precedent, build 2–3 competing prototypes and compare side by side before committing.
- [principle-build-the-lever](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-build-the-lever/SKILL.md) — core. For any non-trivial work, build the tool that does or proves it (codemod, script, generator, skill) instead of working by hand. The tool is what a reviewer reruns.
- [principle-model-the-domain](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-model-the-domain/SKILL.md) — architecture. Encode the domain in a structure (state machine, typed model, registry, reducer) instead of scattered conditionals.
- [principle-boundary-discipline](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-boundary-discipline/SKILL.md) — architecture. Concentrate guards at system boundaries (CLI, config, network, external APIs); trust internal types and keep business logic in pure functions.
- [principle-type-system-discipline](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-type-system-discipline/SKILL.md) — architecture. Make illegal states unrepresentable, brand semantic primitives, parse external data at boundaries, never lie to the compiler, exhaust variants, derive from authoritative schemas.
- [principle-make-operations-idempotent](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-make-operations-idempotent/SKILL.md) — architecture. Commands, lifecycle steps and loops converge to the same end state regardless of partial prior runs, crashes and retries.
- [principle-migrate-callers-then-delete-legacy-apis](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-migrate-callers-then-delete-legacy-apis/SKILL.md) — architecture. When a new internal API replaces an old one, migrate every caller and delete the old API in the same wave.
- [principle-separate-before-serializing-shared-state](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-separate-before-serializing-shared-state/SKILL.md) — architecture. When concurrent actors might write the same file, branch, key or object, eliminate the sharing first; serialize only when one shared writer is a real invariant.
- [principle-prove-it-works](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-prove-it-works/SKILL.md) — verification. Before declaring done, verify against the real artifact (run the feature, read the actual value, inspect the diff), not a proxy, self-report or "it compiles".
- [principle-fix-root-causes](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-fix-root-causes/SKILL.md) — verification. Reproduce first, ask why until you reach the root cause and fix it there; resist nil-check guards that silence crashes.
- [principle-sequence-verifiable-units](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-sequence-verifiable-units/SKILL.md) — verification. Break multi-step work into small units that each end in a verifiable state, check each before the next, and order commits so the sequence proves itself.
- [principle-test-behavior-not-implementation](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-test-behavior-not-implementation/SKILL.md) — verification. Call the code the way users do and assert what they observe against a literal. If the test still passes when every import returns undefined, rewrite or delete it.
- [principle-guard-the-context-window](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-guard-the-context-window/SKILL.md) — delegation. Route bulk (large outputs, long files, fan-out) to subagents; keep summaries in the main thread, not raw payloads.
- [principle-never-block-on-the-human](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-never-block-on-the-human/SKILL.md) — delegation. On reversible work, proceed, present the result and let the human course-correct; reserve confirmation for irreversible actions.
- [principle-encode-lessons-in-structure](https://raw.githubusercontent.com/cursor/plugins/main/pstack/skills/principle-encode-lessons-in-structure/SKILL.md) — meta. When you write the same instruction twice, encode it as a lint, metadata flag, runtime check or script instead of more text.

## poteto-mode playbooks

- [Investigation](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/investigation.md) — A read-only question: how does X work, why was Y built this way, are we sure.
- [Bug fix](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/bug-fix.md) — Reproduce a defect, root-cause it, and fix with runtime evidence.
- [Perf issue](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/perf-issue.md) — Trace a measured slowness and improve it against a baseline.
- [Hillclimb](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/hillclimb.md) — Sustained improvement of one metric against a target: hypothesis loop, before/after measurement, one commit per accepted win.
- [Runtime forensics](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/runtime-forensics.md) — Diagnose a live symptom (leak, idle-CPU spin, glitch) from instrumentation.
- [Trace forensics](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/trace-forensics.md) — Diagnose a captured profile (cpuprofile, trace, spindump, heap snapshot).
- [Feature](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/feature.md) — New or changed behavior, built from a named data shape.
- [Refactoring](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/refactoring.md) — A behavior-preserving change to structure or shape.
- [Prototype](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/prototype.md) — A throwaway sketch to settle a design or empirical fork by observing it instead of asking.
- [Visual parity](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/visual-parity.md) — Pixel-exact UI equivalence between two implementations.
- [Authoring a skill](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/authoring-a-skill.md) — Writing or editing a SKILL.md.
- [Eval](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/eval.md) — Test, blinded, how a skill or prompt change affects agent behavior.
- [Babysit](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/babysit.md) — Drive a PR or stack to merge-ready: conflicts, review threads, CI.
- [Shipping](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/shipping.md) — Independently verify a green stack, then land the contiguous verified run bottom-up.
- [Autonomous run](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/autonomous-run.md) — Drive a long task to completion without stopping (pairs with /loop).
- [Orchestrate](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/orchestrate.md) — A standing, multi-day project for one coordinator: many stacked PRs, fleets of subagents.
- [Autopilot-full](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/autopilot-full.md) — Run independent PRs to merged, one owner per PR, a root swarm verdict each round.
- [Autopilot-stack](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/autopilot-stack.md) — Build and verify one linear stack for the operator to review and land.
- [Session pickup](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/session-pickup.md) — Resume or take over a prior agent's in-flight work.
- [Pause safely](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/pause-safely.md) — Suspend in-flight work cleanly so it can be resumed.
- [Multi-phase plan](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/multi-phase-plan.md) — Work that spans phases or stacked PRs.
- [Worktree cleanup](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/worktree-cleanup.md) — Reclaim disk by pruning merged or abandoned worktrees, safety-gated.
- [Opening a PR](https://github.com/cursor/plugins/blob/main/pstack/skills/poteto-mode/playbooks/opening-a-pr.md) — Small ordered commits, conventional title, briefing-style body. Ends every other playbook.

## Agents

- poteto-agent (agents/poteto-agent.md) — The subagent every playbook step spawns. Reads poteto-mode in full, principles index included, before doing anything. Copy to `~/.claude/agents/`; spawn with the Agent tool's `subagent_type: "poteto-agent"`. Cursor's `is_background` → Claude Code `background: true`.
- Comment Sicko (agents/comment-sicko.md) — A read-only comment reviewer that hunts narration, banners and workaround comments. Invoked through /no-comments. Copy to `~/.claude/agents/`.

## Benny automation pack (not slash skills)

- setup-benny (automations/benny/skills/setup-benny/SKILL.md) — Configures the triage and repro automations: Slack channel, tracker, repository, routing, control adapter, models, budget. Cursor Automations only.
- triage-issue-reports (automations/benny/skills/triage-issue-reports/SKILL.md) — One thread-only verdict per Slack report: classify, trace the owning layer, dedupe against the tracker, create a ticket only for a clear new bug. Run from a GitHub Action or a scheduled agent instead of a Cursor Automation.
- reproduce-and-fix-issues (automations/benny/skills/reproduce-and-fix-issues/SKILL.md) — Reproduces a triaged bug twice through the real UI, verifies existing fixes, and opens one bounded draft PR only after before-and-after proof. Same: hosted trigger + Playwright MCP as the control adapter.

## Cursor → Claude Code

- /add-plugin pstack → Copy skills/ and agents/ into ~/.claude/ (install script below)
- ~/.cursor/rules/pstack-models.mdc (always-applied rule) → A pstack section in ~/.claude/CLAUDE.md (always loaded)
- Task tool → Agent tool
- subagent_type: generalPurpose → subagent_type: general-purpose
- model: grok-… / gpt-… / claude-…-max → model: fable | opus | sonnet | haiku (Claude models only)
- AskQuestion → AskUserQuestion
- ~/.cursor/projects/ (transcripts) → ~/.claude/projects/<project>/*.jsonl
- .cursor/skills/verify-<app>/ → .claude/skills/verify-<app>/
- create-skill (Cursor built-in) → Edit SKILL.md directly, or the skill-creator skill
- /deslop (cursor-team-kit) → /simplify, then /code-review
- control-ui (cursor-team-kit) → Playwright MCP (browser_navigate, browser_click, …)
- control-cli (cursor-team-kit) → Bash, or a PTY via expect/tmux
- /loop (Cursor built-in) → /loop (Claude Code built-in)
- mode: true (sticky mode) → Not supported — re-invoke /poteto-mode or say "stay in poteto-mode"
- Isolated worktree per candidate → Agent tool isolation: "worktree"