hafuture

mattpocock/skills vs pstack: Beginner to Intermediate Guide

Compare mattpocock/skills and pstack side by side: 7-step conversation guardrails for beginners vs. parallel agents and review panels for intermediate devs.

AIVibe Codingmattpocock/skillspstackCursorClaude CodeCavemanPonytail

Table of Contents


1. The One-Line Verdict


2. Two Analogies That Make It Click

  • mattpocock/skills = A Question-Happy Private Tutor Starts by asking: "Tell me exactly what we're building first." It turns your answers into a spec, slices that spec into ordered tickets, and prevents wrong turns before a single line of bad code gets written.
  • pstack = A Strict Site Foreman You toss a task into /poteto-mode, the foreman picks the right playbook out of 23, runs it with strict engineering rules, and has a multi-model review panel verify the output. Its slogan says it all: "fearless parallelism."

Their creators' backgrounds mirror those personalities. TypeScript educator Matt Pocock notes that agents usually fail not from syntax mistakes, but from guessing what you meant—hence his question → spec → ticket sequence. Meanwhile, poteto (formerly Meta, Netflix, and Cursor, working on the React Compiler) built pstack to stop AI "slop" and ship less, higher-quality code.


3. Side-by-Side Comparison

Dimensionmattpocock/skillspstack
CreatorMatt Pocock (Total TypeScript)poteto (Meta, Netflix, Cursor, React Core Team)
Home AgentAgent-agnostic (Claude Code, Codex, Copilot, Gemini, Cursor)Cursor first (supports Claude Code with /pstack: prefix & manual auto-update toggle)
Scope27 skills (16 user-invoked + 11 model-invoked)23 playbooks + skills + 2 subagents + 24 principles
Main Entry7-step recipe: /grill-with-docs → /to-spec → /to-tickets → /implement → /code-review → /pr → /retroSingle gateway: /poteto-mode (auto-matches the right playbook)
What It EnforcesQuestion → spec → ticket order (dialogue habits)Rigor, verification & parallelism (execution quality)
Setup30 seconds (plugin or npx skills)1 line in Cursor (/add-plugin pstack), marketplace add in Claude Code
Best Fit For"My AI keeps building the wrong thing!""I want to delegate large tasks end-to-end safely."
Peak MomentAligning direction before coding, stopping recurring bugsTough debugging, performance fixes, overnight autonomous runs

4. Why Beginners Thrive with mattpocock/skills

1. The Question Barrage Teaches You Product Design

Before writing code, /grill-with-docs asks concrete questions—"Where is data stored? Do we need auth? What happens when a job fails?"—catching planning blind spots beginners often overlook. It doubles as a hands-on engineering coach while building GLOSSARY.md and ADRs that shrink future prompts.

2. Only One 7-Step Sequence to Remember

While pstack enters through /poteto-mode, behind that door sit reasoning-budget setups (/setup-pstack), 7 playbook categories, and autonomous workflows. With mattpocock/skills, beginners never have to guess what to invoke next—just follow the 7-step recipe, or run /ask-matt whenever stuck.

3. Works Out of the Box on Any Agent

If you already use Claude Code, Codex, GitHub Copilot, or Gemini CLI, setup takes 30 seconds with auto-updates built in. Because pstack originated in Cursor, using it inside Claude Code requires knowing extra adapter details: commands get prefixed (/pstack:poteto-mode), model mappings live in ~/.claude/rules/pstack-models.md, and third-party marketplace auto-updates must be enabled manually.


5. Why Intermediate Devs Thrive with pstack

1. True Multi-Agent Parallelism

Where mattpocock/skills sharpens a single session's workflow, pstack is explicitly engineered to orchestrate multiple agents simultaneously—splitting tasks into verifiable units and merging them cleanly (one real example prompt literally reads: "I'm going to sleep. Merge everything by morning.").

2. Diagnostic Playbooks for Tough Bugs

For elusive regressions—"I fixed it twice and it still drifts every 750ms"—pstack ships dedicated playbooks that enforce reproduce → diagnose → fix → verify, attacking the exact debugging and performance bottlenecks where intermediate engineers lose the most hours.

3. Multi-Model Review Panels

Having multiple models cross-examine changes catches subtle bugs a single pass misses. It deliberately trades extra tokens for higher reliability—worth every penny when rewriting broken code costs far more than API tokens.


6. Quick Situation Picker

Which Setup Matches Where You Are Today?
mattpocock/skills

The question-first tutor for clean specs & tickets

“/grill-with-docs → /to-spec → /implement”

Core Strength
Closes communication gaps early and locks domain terms in GLOSSARY.md
Ideal Agents
Claude Code, Codex, Copilot, Gemini CLI, Cursor
Learning Curve
Low (follow the 7-step recipe or run /ask-matt)
pstack

The strict site foreman with 23 playbooks & review panels

“Fearless parallelism”

Core Strength
Repro-first debugging, multi-model panels, and parallel execution
Ideal Agents
Cursor native (Claude Code supported with /pstack: prefix)
Learning Curve
Intermediate (shines when you know how to verify output)
1 Process Tool + Token Savers

Don't run both process frameworks—layer Caveman & Ponytail instead

“One conductor, zero rule collisions”

Why Not Both?
Mixing 'grill first' and 'follow playbook' forces constant routing overhead and doubles rule tokens
Clean Fix
Keep either matt OR pstack, then layer Caveman + Ponytail underneath
Golden Rule
If total billed cost rises on the same task tier, turn the saver combo off

7. Should You Use Both? (4-Step Cleanup & Token-Saver Stack)

Technically you can install both, but you shouldn't. Because both enforce structured procedures, their conductors collide—forcing your agent to decide every turn whether to follow Matt's 7-step recipe or pstack's playbook while doubling rule-file bloat.

Instead, follow "One Process Tool, Optional Savings Layers":

bash
claude plugin install mattpocock-skills@claude-plugins-official
# Then run once per repository:
/setup-matt-pocock-skills
Leaves mattpocock/skills as the single conductor for your repo.

Step 4: Verify Single Conductor & Measure Real Billing Impact

Never judge savings solely by shorter chat replies—rule prompts consume input tokens too. Compare total session cost on your provider's billing/usage page:

  1. 1
    Pick two small tasks of similar difficulty

    Don't repeat the exact same task twice (the second run benefits from already knowing the answer). Pick two comparable small fixes or helper functions.

  2. 2
    Run #1 with Caveman off and log the billcaveman off

    Disable Caveman, run claude as usual on task #1, and immediately record tokens and cost on your provider's usage dashboard.

  3. 3
    Run #2 with Caveman enabled and log the billcaveman trial -- claude

    Complete task #2 under caveman trial -- claude and check the exact same usage dashboard right afterward.

  4. 4
    Keep only if total cost actually droppedcaveman stats

    If run #2 cost less, keep it on (and track history via caveman stats). If total cost matched or rose, turn it off for that workload.

Click any verification command to copy

Verification & A/B Measurement Commands


8. Growth Roadmap & Post-Setup Checklist

Open a fresh session after setup and verify these three items. If anything looks off, revisit Step 1 ("one conductor only"):

CheckIf You Kept mattpocock/skillsIf You Kept pstack
1. Single command set/ask-matt and /grill-with-docs work; /pstack:* does not appear/pstack:poteto-mode works; /grill-with-docs does not appear
2. No duplicate rule filesCLAUDE.md / AGENTS.md contains rules for one workflow tool only (mixing both doubles context overhead)Same—confirm no leftover mattpocock workflow rules remain
3. Token-saver checkRun a real task with Caveman on and verify net bill reduction via caveman stats & usage pageSame—turn off Caveman if total billed cost rises

9. Frequently Asked Questions (FAQ)

Q1. Can a beginner start directly with pstack?
Yes—calling /poteto-mode alone is simple enough. However, pstack's real strengths (multi-model review panels, parallel agents, and overnight runs) only pay off once you can critically verify agent output. Beginners usually learn faster and waste fewer tokens by starting with mattpocock/skills' 7-step conversation guardrails.
Q2. Can I install and run mattpocock/skills and pstack together?
Technically yes, but it is not recommended. Both tools enforce end-to-end procedures, so their conductors collide and rule files bloat twice as fast. Keep only one process tool, and pair it with non-overlapping token savers like Caveman and Ponytail instead.
Q3. Does pstack work in Claude Code, or only in Cursor?
It works in Claude Code via an adapter, though Cursor is its native home. In Claude Code, commands get prefixed (e.g., /pstack:poteto-mode), auto-update must be enabled manually under /plugin > Marketplaces > cursor-plugins, and Cursor-only steps like bugbot or automations are skipped with a short note.
Q4. Are both tools free, and which consumes more tokens?
Both are free, open-source repositories (you only pay your existing AI subscription or API usage). Because pstack uses multi-model review panels and parallel execution, it deliberately trades extra tokens for higher quality. If cost matters, layer Caveman underneath and compare real totals on your provider's billing page using caveman trial -- claude.
Q5. How do I quickly confirm there are no duplicate rules or command conflicts?
Open a fresh session and check three things: (1) only your chosen tool's commands appear (/ask-matt OR /pstack:poteto-mode), (2) CLAUDE.md and AGENTS.md contain workflow instructions from just one tool, and (3) caveman stats plus your billing dashboard confirm total session cost actually decreased.

Slim Down CLAUDE.md & AGENTS.md Rules Locally with Hafuture

Strip extra line breaks and messy whitespace when merging rule files in your browser—100% private with zero server uploads.

Open Whitespace Remover