pstack is a Cursor plugin of ~45 skills built around one idea: an orchestrator mode
(/poteto-mode) reads your task, picks a playbook, and routes to specialized skills and
21 engineering principles as steps fire. This document explains what each skill does and how they
fit together.
Three skills define how work gets shaped. Everything else is a step they call.
/poteto-mode orchestratorThe entry point and sticky mode. Opens a todo list whose first item is reading its inline principles index, matches your request to one of 22 playbooks (bug fix, perf, feature, shipping, autonomous run, orchestrate…), copies the steps verbatim, and invokes the other skills as those steps fire. Stays on across turns; opts out only when you say so.
/setup-pstack Detects which models you can access and writes an always-applied rule mapping roles (precise code → sol, mechanical code → grok, prose/judgment → fable…) to models. Every multi-model skill reads this rule and falls back to defaults if absent.
/automate-me Mines your recent transcripts and drafts a personal <your-name>-mode skill
from how you actually work, routing through pstack underneath. Uses Cursor's built-in
create-skill plus unslop.
/figure-it-out The escape hatch when no bundled playbook fits: designs a rigorous, auditable playbook for the task at hand, scales rigor to the stakes, runs a hypothesis loop.
/how understandWalkthroughs of subsystem architecture and runtime flow; placement/ownership questions ("where should this live"). Evidence comes from the code itself.
/why understandDesign rationale and history. Discovers available MCPs at runtime and queries each evidence category in parallel: source control, issue tracker, long-form docs, chat, infra observability, error tracking, analytics.
/recall understandRebuilds recent working context from your own chat history plus the shared record and hands back a tight current-state brief. Typically run before starting or resuming a task — often the first thing poteto-mode does on session pickup.
/teach understandA composite: runs how + why and weaves both findings into one plain
explanation, built up diagram by diagram. Use when you want genuine understanding, not a summary.
/blast-radius understandFor small-looking changes: finds what else they could break, and proves the one fact you're safe because of by running real code — not by asserting it.
/architect buildBefore code crosses a function boundary: settle the caller's usage, types, and module shape first, then stay in the loop while implementation fills in. Prevents jumping to code that locks in the wrong shape.
/tdd buildFailing test first, then the fix — only when there's a cheap local test path. Pairs with the fix-root-causes principle (reproduce first).
/typescript-best-practices buildGrounds the type-system-discipline principle in concrete TS syntax. Auto-applies whenever a
.ts/.tsx file is touched.
/arena parallelN parallel attempts at the same task across different models, then pick a base and graft the strongest parts of the losers into it. Best-of-N with recombination.
/swarm parallelN parallel workers across different slices — coverage matrices, races, gauntlets, exploration — drained into one aggregated report. Embodies guard-the-context-window: bulk stays in children, summaries come back.
/interrogate parallelMulti-model adversarial review: several LLM reviewers try to break a diff from independent angles, including a strict code-quality lens. The last gate before shipping contested designs.
/create-verification-skill verifyIf a project has no scripted way to prove app behavior, generates a project-local verify skill with a feature map — drives the app the way a user would, any language or platform.
/maintain-verification-skill verifyKeeps that verify skill honest over time: parallel source readers per feature plus one live pass driving every feature; at most one PR of proven corrections.
They industrialize the prove-it-works principle ("verify against the real
artifact, not a proxy"). Playbooks like shipping and autonomous-run lean on the generated verify
skill so agents can prove their own work without a human in the loop.
/unslop proseRemoves AI writing tells. Applies to every reply poteto-mode writes ("any prose surface"), and to code comments and docs.
/technical-writing proseLayered doc standard (Diátaxis + Google developer style + STE + Global English) for docs, RFCs, readmes, PR descriptions, and commit messages.
/no-comments proseStrips comments before review: spawns the Comment Sicko subagent (read-only),
fixes accepted findings, offers encodings for claimed constraints. Runs before review; complements
the external /deslop which runs before commit.
/bro proseRestates the last message in plain human language, no jargon. A one-shot readability valve.
/show-me-your-work recordA TSV decision trail (what / why / evidence / result) for long-running or unattended work. Local by default; committed when a reviewer needs to trust the result.
/reflect After a long task lands: spawns three parallel review subagents over the transcript, surfaces learnings, and routes each one to a concrete edit on an existing skill.
One short leaf skill per principle. poteto-mode carries an inline index of them and requires each applied principle to trace to a concrete decision. Standalone files exist so other skills can reference principles by name.
| Group | Principles |
|---|---|
| Core ×9 | laziness-protocol · foundational-thinking · redesign-from-first-principles · subtract-before-you-add · minimize-reader-load · outcome-oriented-execution · experience-first · exhaust-the-design-space · build-the-lever |
| Architecture ×6 | model-the-domain · boundary-discipline · type-system-discipline · make-operations-idempotent · migrate-callers-then-delete-legacy-apis · separate-before-serializing-shared-state |
| Verification ×3 | prove-it-works · fix-root-causes · sequence-verifiable-units |
| Delegation ×2 | guard-the-context-window · never-block-on-the-human |
| Meta ×1 | encode-lessons-in-structure |
typescript-best-practices implements type-system-discipline;
tdd operationalizes fix-root-causes;
swarm embodies guard-the-context-window;
create-verification-skill industrializes prove-it-works;
reflect enacts encode-lessons-in-structure;
arena executes exhaust-the-design-space.Subagent (subagent_type: "poteto-agent") that runs poteto's style end-to-end.
Reads poteto-mode in full — including the principles index — before doing any work. Same wrapper
poteto-mode routes through.
Read-only comment-reviewer subagent. Almost always invoked indirectly via /no-comments, not directly.
Dormant pack under automations/benny/: triages Slack issue reports, reproduces
confirmed bugs, fixes with real UI evidence. Not registered as slash skills; enabled per-repo via FOR_AGENTS.md.
Ships /deslop (pre-commit) and control-cli / control-ui
(shipping CLIs/UIs). Referenced by poteto-mode but installed separately.
Plan mode replaces planning skills; create-skill is used by automate-me and authoring
playbooks; the built-in babysit is deliberately superseded by poteto-mode's babysit playbook.
/poteto-mode scroll drifts every 750ms, repro first, fix, verify. Mode opens its todo list, reads the principles index, matches the bug fix playbook. Model roles resolve from the rule written by setup-pstack.| Skill | Does | Relationships |
|---|---|---|
| ORCHESTRATION & META | ||
poteto-mode | Sticky orchestrator; 22 playbooks; routes all other skills | Hub — indexes all principles, invokes nearly every skill below |
setup-pstack | Maps model roles to your available models via a config rule | Read by every multi-model skill (arena, swarm, interrogate, reflect…) |
automate-me | Drafts a personal "-mode" skill from your transcripts | Uses create-skill + unslop; keeps pstack as base |
figure-it-out | Designs auditable ad-hoc playbooks when none fits | Logs via show-me-your-work; alternative to bundled playbooks |
| UNDERSTANDING | ||
how | Subsystem walkthroughs; placement/ownership questions | Composed by teach; cited by poteto-mode triggers |
why | Design rationale via parallel MCP evidence queries | Composed by teach |
recall | Rebuilds recent context into a current-state brief | Session pickup / resume entry point |
teach | Weaves how + why into one plain explanation with diagrams | Composite of how + why |
blast-radius | Finds what a change could break; proves safety by running code | Bridges understanding → verification |
| BUILDING & PARALLELISM | ||
architect | Settles types/signatures/module shape before coding | Gate before implementation in feature/refactor playbooks |
tdd | Failing test first, then fix | Operationalizes fix-root-causes |
typescript-best-practices | TS-specific rules when reading/editing .ts files | Grounds type-system-discipline in syntax |
arena | N attempts at same task; pick base, graft best parts | Executes exhaust-the-design-space; contrast with swarm |
swarm | N workers on different slices; one aggregate report | Embodies guard-the-context-window |
interrogate | Multi-model adversarial review of a diff | Final gate after arena/swarm produce |
| VERIFICATION | ||
create-verification-skill | Generates project-local verify skill + feature map | Industrializes prove-it-works; consumed by shipping/autonomous playbooks |
maintain-verification-skill | Keeps verify skill honest vs. the app | Maintains create-verification-skill's artifact |
| PROSE, RECORD & LEARNING | ||
unslop | Removes AI writing tells from any surface | Applies to every poteto-mode reply; used by automate-me |
technical-writing | Diátaxis + Google + STE + Global English standard | Docs, RFCs, PRs, commit messages |
no-comments | Comment-free diffs before review | Spawns Comment Sicko; pairs with external deslop (pre-commit) |
bro | Restates last message in plain language | Standalone readability valve |
show-me-your-work | TSV decision trail for unattended work | Consumed by figure-it-out; written during autonomous runs |
reflect | Mines transcript; routes learnings into skill edits | Enacts encode-lessons-in-structure; closes the loop |
| PRINCIPLES (21 leaf skills) | ||
principle-* | One rule each: core ×9, architecture ×6, verification ×3, delegation ×2, meta ×1 | Indexed inline by poteto-mode; referenced by name from other skills |