Best Claude Code skills and plugins for product managers (2026)

The short answer. The best Claude Code skills for product managers in 2026 are Anthropic's free Product Management plugin (specs, roadmaps, stakeholder updates, research synthesis), the free community pack phuryn/pm-skills, and Evals Skills if you own an AI feature. Add a focused plugin such as Porter strategy skills, PMF Skills or Where Did We Land? when you need one job done well. If you want context, workflows and review agents already wired together, AI PM OS is the paid option ($99 one-time). Add design plugins such as Taste Skill once you start prototyping.

Start here: Anthropic Product Management plugin

Terminal
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install product-management@knowledge-work-plugins
Jump to the picks

Full disclosure: AI PM OS is prodmgmt.world's own product, and I built the free Porter, PMF, Where Did We Land?, Tufte, Refactoring UI and nurijanian-skills entries. I've listed AI PM OS after the free PM packs and described the others using their makers' published facts.

By George Nurijanian24 min read

Quick picks

Write specs and PRDsFree
Anthropic PM plugin (/write-spec) or phuryn /write-prd
Run discovery and research synthesisFree
phuryn /discover, or Anthropic /synthesize-research
Do strategy and prioritizationFree
phuryn /strategy, or Porter strategy skills /analyze-five-forces for competitive analysis
Test a new product idea for product-market fitFree
PMF Skills /pmf-hypothesis-builder
Turn a meeting transcript into what was decidedFree
Where Did We Land?
Judge whether your AI feature worksFree
Evals Skills
Prototype UI that doesn't look AI-builtFree
Taste Skill
Make charts that don't misleadFree
Tufte VDQI
Stop Claude over-buildingFree
Ponytail, plus Matt Pocock's /grill-me before you build
Have context, workflows and reviewers already wired$99 one-time
AI PM OS

How to choose, and how many to install

A stack that works for most PMs:

01

PM skill packs

One PM pack. Anthropic's plugin or phuryn's if you want commands, plus a focused plugin such as Porter or PMF Skills for one job. AI PM OS if you want the system assembled.

Pick · Product Management

1

Anthropic Product Management plugin

Claude Code plugin

Official · free · Apache 2.0 repo

The default first install, especially if your team uses Cowork.

Anthropic's own PM plugin, part of its knowledge-work-plugins collection. It's designed primarily for Cowork and also works in Claude Code. Seven commands: /write-spec, /roadmap-update, /stakeholder-update, /synthesize-research, /competitive-brief, /metrics-review and /brainstorm, backed by seven skills (feature spec, roadmap management, stakeholder comms, research synthesis, competitive analysis, metrics tracking, brainstorming).

How PMs use it

Connect Linear or Jira, Slack and Notion, run /stakeholder-update on Friday, and it pulls the week's tickets and threads into a draft for your exec audience.

Pros

  • Maintained by Anthropic
  • Ships MCP connectors for trackers, Slack, Notion, Figma, Amplitude and Intercom
  • Installs from Cowork's plugin screen

Cons

  • Seven commands, so narrower than the community packs
  • Drafts stay generic until you connect tools or add your own context

Install

Terminal
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install product-management@knowledge-work-plugins
Plugin page

Pick · Product Management

2

PM Skills Marketplace (phuryn/pm-skills)

Claude Code plugin

Free · MIT · ~27.3k stars

The free pick if you want ready commands for most of the PM job.

Paweł Huryn's pack. The README lists 69 PM skills and 42 chained workflows across 9 plugins, covering discovery, strategy, execution, market research, analytics, go-to-market, growth and shipping AI-built code. Start with /discover, /strategy or /write-prd. /red-team-prd stress-tests a PRD and ranks the riskiest assumptions; /pre-mortem runs a pre-mortem on a PRD or launch plan.

Pros

  • The broadest free set
  • Commands chain several skills
  • Installs in Cowork from the GitHub marketplace option; Codex reads the same marketplace

Cons

  • You supply product context
  • Slash commands are Claude-specific (in Codex and Cursor you get the skills, not the commands)

Install

Terminal

Adds the three plugins behind /discover, /strategy and /write-prd. The other six install the same way.

claude plugin marketplace add phuryn/pm-skills
claude plugin install pm-product-discovery@pm-skills
claude plugin install pm-product-strategy@pm-skills
claude plugin install pm-execution@pm-skills
Repo
3

Evals Skills

Skill pack, not a plugin

Free · Apache 2.0 · 1.5k+ stars

Product-specific AI evals from the AI Evals course

Show details

Evals Skills come from Hamel Husain and Shreya Shankar, who teach the AI Evals course and have helped 50+ companies with evals. An evals-start skill looks at your situation and routes you. eval-audit inspects an existing eval pipeline and ranks its problems by severity. error-discovery builds a review app for your traces, picks a diverse sample, and sorts your notes into failure modes. Other skills generate synthetic test data, write code checks and LLM judges, and calibrate those judges against human labels.

How PMs use it

If you own an AI feature, "it seems fine" is not a quality bar. Point error-discovery at a file of real outputs and you read and annotate them in a simple review app while the agent groups your notes into failure modes. That gives you a ranked list of what is actually going wrong before anyone writes an eval, which is the step most teams skip.

Pros

  • Built by the people who teach the most-used AI evals course
  • Starts with error analysis, the step most teams skip
  • Review app is a single HTML file with no dependencies
  • Covers LLM judges, synthetic data, and RAG evaluation

Cons

  • Only useful once you have an AI feature with real outputs
  • Some skills assume comfort with data files and Python
  • The skills avoid common mistakes but do not replace the course

Install

Terminal
npx skills add https://github.com/ai-evals-course/evals-skills
Agent chat

Then point the agent at your data:

"Can you help me do error analysis on traces.jsonl?"

Verdict: The best PM-facing skill pack for AI features. Install it the week your team ships anything built on an LLM.

Docs and GitHub
4

NotebookLM MCP CLI

MCP server and CLI, not a plugin

Free · MIT · 6.3k+ stars

Drive NotebookLM from Claude: sources, questions, and audio overviews

Show details

NotebookLM MCP CLI (by Jacob BD) connects Claude and other agents to Google's NotebookLM, which Google now calls Gemini Notebook. It ships an nlm command line tool and an MCP server with 53 tools. Your agent can create notebooks, add URLs, Drive files, and documents as sources, ask questions across them, and generate audio overviews, videos, and slide decks. A setup wizard finds the AI tools on your machine and connects them.

How PMs use it

Research synthesis is where NotebookLM shines, but loading sources by hand is slow. Ask Claude to create a notebook from your interview notes, competitor pages, and support tickets, then query it for patterns with citations back to the source. Generate an audio overview for stakeholders who will not read the doc.

Pros

  • Turns NotebookLM into a research tool your agent can drive
  • Answers cite the sources they came from
  • Audio overviews, videos, and slides for stakeholders
  • Setup wizard connects Claude Code, Codex, and Cursor

Cons

  • Uses Google's internal, undocumented APIs that can change without notice
  • May not align with Google's Terms of Service, so keep it to personal use
  • Login relies on cookies taken from your browser

Install

Terminal
uv tool install notebooklm-mcp-cli
nlm login
Terminal

Connect Claude Code and your other AI tools:

nlm setup

Verdict: The fastest way to turn a pile of research into cited answers. Use it for personal research, not for anything that has to be reliable.

Docs and GitHub
5

Porter Competitive Strategy Skills

Claude Code plugin

Free · CC BY 4.0 · ~61 stars

Pick it when you need competitive strategy done properly, not a generic SWOT.

Show details

12 skills and an orchestrator extracted from Michael Porter's Competitive Strategy: five forces, competitor profiling, strategic groups, market signals, generic strategies, competitive moves, market entry, and playbooks for fragmented, emerging and declining industries. Run /orchestrate-porter-strategy and it routes your question to the right skill, or call one directly, such as /analyze-five-forces or /profile-competitor. (Full disclosure: I built this one.)

How PMs use it

Before a strategy review, ask "analyze this industry" for a five forces assessment, or "what move should we make?" for a competitive move with a retaliation forecast. The output follows Porter's own checklists, so you can defend it in the room.

Pros

  • Grounded in Porter's original frameworks, not summaries
  • An orchestrator picks the skill, so you can just ask
  • Skills chain: five forces output feeds the move and entry skills

Cons

  • Industry-level strategy, not day-to-day product decisions
  • Smaller community than the big PM packs

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add gnurio/porter-strategy-skills
  2. /plugin install porter-strategy-skills
Claude Code chat

Then ask a strategy question, or start with:

/orchestrate-porter-strategy

Verdict: Pick it when you need competitive strategy done properly, not a generic SWOT.

Repo
6

PMF Skills

Cursor plugin

Free · CC BY 4.0 · ~28 stars

Pick it when you are testing a new product idea and need evidence before you build.

Show details

A five-skill workflow for product-market fit discovery across six dimensions: problem, target audience, value proposition, competitive advantage, growth strategy and business model. /pmf-hypothesis-builder writes your V1 hypothesis, /pmf-market-research sends parallel agents for analogs and failure cases, /pmf-research-synthesis scores risk, /pmf-validate designs an experiment for the riskiest assumption, and /pmf-status tells you where you are. (Full disclosure: I built this one.)

How PMs use it

Run /pmf-hypothesis-builder on a new product idea and answer its questions. You end with a written hypothesis, a ranked risk table and one specific next experiment, not "do more research".

Pros

  • One clear path from raw idea to an experiment brief
  • /pmf-status is read-only, so it is safe to run after a break
  • Plugin in Cursor; plain skills in Claude Code and Codex

Cons

  • Built for new products and bets, not mature features
  • The full run takes a couple of hours across sessions

Install

Cursor chat
/add-plugin https://github.com/gnurio/pmf-plugin
Terminal

Claude Code or Codex: clone into your project, then run /pmf-hypothesis-builder

git clone https://github.com/gnurio/pmf-plugin

Verdict: Pick it when you are testing a new product idea and need evidence before you build.

Repo
7

Where Did We Land?

Claude Code plugin

Free · MIT · ~40 stars

Install it if you run meetings where decisions get lost.

Show details

Turns a meeting transcript into one self-contained HTML page: a topic timeline, a ledger of where every thread landed with the quote that proves it, open loops, commitments, stances and airtime. It reads Zoom, Teams, Otter, Granola, Fathom and other transcript formats, and if you give it nothing it looks for the latest transcript on your machine. (Full disclosure: I built this one.)

How PMs use it

After a decision meeting, run /where-did-we-land on the transcript and send the page round. Every "we agreed" comes with the quote, so nobody re-litigates it next week.

Pros

  • Plain HTML file: no hosting, no server, no dependencies
  • Every landed thread cites the quote that proves it
  • Works with the transcript formats most meeting tools export

Cons

  • Only as good as the transcript you feed it
  • One meeting at a time, not a running decision log

Install

Terminal

Any agent:

npx skills add gnurio/where-did-we-land
Claude Code chat· one prompt each

Or as a Claude Code plugin:

  1. /plugin marketplace add gnurio/where-did-we-land
  2. /plugin install where-did-we-land

Verdict: Install it if you run meetings where decisions get lost.

Repo
9

nurijanian-skills (free sample)

Skill pack, not a plugin

Free · MIT · ~127 stars

A no-cost way to judge the quality bar.

Show details

My free, smaller set of PM skills, written to the same standard as AI PM OS. Start with make-requirements-great.

Install

Terminal

installs to Claude Code, Cursor and Codex

npx nurijanian-skills
Terminal

Claude Code only

npx nurijanian-skills --claude

Verdict: A no-cost way to judge the quality bar.

Repo

Don't stack several PM packs at once. Many share command names like /write-prd or /brainstorm, and you'll lose track of which one ran. Try one for two weeks, then add or swap.

02

Your own context file.

Whatever pack you pick drafts generic output until Claude knows your product. Write a CLAUDE.md or AGENTS.md (how).

03

Design plugins for PMs who prototype

One design plugin if you prototype. Claude writes working UI by default, not good-looking UI. Install one of these once you start building prototypes (prototype guide).

Pick · Design

10

Taste Skill

Claude Code plugin

101k+ stars

Install it before your next prototype. It is the fastest fix for AI UI that looks like AI UI.

Taste Skill is a set of portable agent skills that push AI-built interfaces toward stronger layout, typography, motion, and spacing. The default skill reads your brief, infers a design language, and tunes three dials: variance, motion, and density. Variants cover redesigning an existing project, calm premium UI, minimalist editorial UI, and brutalist layouts, plus image-generation skills for reference boards.

How PMs use it

Your prototype is the first thing stakeholders judge, and generic AI styling makes a good idea look half-baked. Install the default skill and Claude stops reaching for the same centered hero and purple gradient. If you already have a product, the redesign skill audits the existing UI first, then fixes layout, spacing, and hierarchy.

Pros

  • Biggest quality jump per minute of setup on this list
  • Works in Claude Code, Cursor, and Codex
  • Pick a style variant once your visual direction is set
  • Redesign skill for existing products, not only greenfield

Cons

  • The default skill is a v2 rewrite still marked experimental
  • Opinionated defaults can fight an established brand system
  • Choosing between ten variants takes a read of the README

Install

Terminal

Default skill:

npx skills add https://github.com/Leonxlnx/taste-skill --skill design-taste-frontend
Terminal

Or install every skill in the pack:

npx skills add https://github.com/Leonxlnx/taste-skill
Claude Code chat· one prompt each

Or as a Claude Code plugin:

  1. /plugin marketplace add Leonxlnx/taste-skill
  2. /plugin install taste-skill@taste-skill
Docs and GitHub
11

Hallmark

Claude Code plugin

31k+ stars

Pick this instead if you mostly build landing pages.

Show details

Hallmark (from Together AI) picks a page structure for your brief, dresses it in one of twenty-one themes, and runs 57 slop-test gates plus a self-critique before it hands anything back. It has four verbs: build new UI, audit existing code against its anti-patterns, redesign a page while keeping the copy and information architecture, and study a screenshot or URL you admire to extract its structure, type pairing, and colour anchor.

How PMs use it

Two pages built for two different briefs should not look like colour swaps of the same template. Run hallmark audit on your current prototype for a punch list of what makes it look AI-built, with no edits. When a competitor's page has the feel you want, hallmark study pulls out the design DNA without copying it pixel for pixel.

Pros

  • Audit mode gives a punch list without touching your code
  • Study mode learns from designs you like and refuses pixel clones
  • Twenty-one themes, so different briefs get different shapes
  • Works in Claude Code, Cursor, and Codex

Cons

  • Overlaps with Taste Skill, so pick one as your default
  • Strongest on marketing and landing pages, less on dense app UI
  • Strong opinions can override an established brand system

Install

Terminal
npx skills add nutlope/hallmark
Claude Code chat· one prompt each

Or as a Claude Code plugin:

  1. /plugin marketplace add nutlope/hallmark
  2. /plugin install hallmark@hallmark
Claude chat· one prompt each
  1. hallmark audit <file or folder>
  2. hallmark study <screenshot or URL>

Verdict: The best pick when the page itself is the product: a launch page, a pitch prototype, or anything stakeholders judge in five seconds.

Docs and GitHub
12

UI UX Pro Max

Claude Code plugin

139k+ stars

Use it to choose a design system when a prototype starts from nothing.

Show details

UI UX Pro Max gives your agent a searchable library of design knowledge: 50 active UI styles, 192 colour palettes matched to product types, 74 font pairings, 25 chart types, and 119 UX guidelines covering accessibility and common production failures. Its design system generator reads your project and recommends a page pattern, style, palette, and type pairing with the reasoning behind each choice. It covers 22 stacks, from React and Tailwind to SwiftUI and Flutter.

How PMs use it

When you start a prototype with no design system, Claude guesses. Describe your product and audience, and UI UX Pro Max proposes a full design system that fits the category, for example calm and trust-heavy for a health app, dense and data-first for a dashboard. The chart recommendations help when your prototype is a metrics view.

Pros

  • Design system recommendations come with reasoning, not just values
  • Industry-specific palettes and patterns for 192 product types
  • Chart guidance for dashboards and analytics prototypes
  • Supports 22 stacks, including mobile

Cons

  • Large knowledge base, so results depend on a clear product brief
  • Some brand and asset features sit in a paid premium version
  • Recommends a style rather than enforcing taste on every screen

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill
  2. /plugin install ui-ux-pro-max@ui-ux-pro-max-skill

Verdict: Use it at the start of a new prototype to pick a design direction that fits the product, then let Taste Skill or Hallmark handle the polish.

Docs and GitHub
13

Design and Refine

Claude Code plugin

780+ stars

Five real variations side by side when you need options to compare.

Show details

Design and Refine (by @0xdesign) generates five distinct variations of any component or page in your actual codebase. It reads your Tailwind config, CSS variables, or component library first, so variations match your design system. You compare them side by side in a design lab route on your dev server, leave Figma-style comments on any element, and it synthesizes a refined version. When you are done it deletes the temp files and writes a DESIGN_PLAN.md.

How PMs use it

Stakeholder alignment on UI is slow when everyone is reacting to descriptions. Run /design-and-refine:start in Claude Code (or /design-lab in Cursor) on the screen you are debating, and you get five real options exploring different layouts, densities, and interaction patterns. Show those instead of arguing about adjectives. The feedback overlay means you can point at the exact button you mean.

Pros

  • Generates real working code, not mockups
  • Infers your existing design tokens before generating
  • Figma-style click-to-comment feedback overlay
  • Cleans up after itself and hands off a DESIGN_PLAN.md

Cons

  • Needs a running dev server (it will not start one for you)
  • Supports React-family frameworks, Vue, and Astro, not everything
  • Five variations per round costs more tokens than one

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add 0xdesign/design-plugin
  2. /plugin install design-and-refine@design-plugins
  3. /design-and-refine:start
Terminal

Cursor or Codex:

npx skills add https://github.com/0xdesign/design-plugin --skill design-lab
Cursor chat
/design-lab

Verdict: The best way to explore a UI direction before committing. Use it when you are stuck between options or need stakeholders to react to something concrete.

Docs and GitHub
14

Layers

Skill pack, not a plugin

300+ stars

When the problem is the product decision beneath the screen.

Show details

Layers (by Jamie Mill) is a product design framework packaged as skills. It splits design into seven layers, from observed behaviour and domain language up through user needs, product strategy, conceptual model, interaction flow, and finally surface. Each /layers-* skill works one layer, and an orient skill tells you which layer your problem actually lives in.

How PMs use it

Most bad screens are bad decisions made one layer down. Before you prototype, run the orient skill on your feature and let it tell you whether the gap is in user needs, the conceptual model, or the flow. It is the closest thing to having a senior product designer ask you the uncomfortable questions first.

Pros

  • Built for product thinking, not pixel pushing
  • Orient skill points you at the right layer to work on
  • Works in Claude Code, Cursor, Codex, and more
  • Pairs well with a UI plugin for the surface layer

Cons

  • Produces decisions and documents, not UI code
  • Seven layers is a framework you have to learn
  • Smaller community than the big skill packs

Install

Terminal
npx skills add jamiemill/layers-skills
Agent chat

Then run any /layers-* skill, starting with:

/layers-orient

Verdict: The most PM-friendly design plugin here. Use it before Taste Skill or Design and Refine, not instead of them.

Docs and GitHub
15

MengTo Skills

Skill pack, not a plugin

6.9k+ stars

A large design-and-build pack; install only the skills you need.

Show details

Meng To's skills library holds 150+ portable SKILL.md playbooks for designers and builders. The flagship workflow turns a screen recording of a design into a detailed recreation prompt, turns existing HTML into reusable interaction prompts, and captures full-page references. The web design set covers landing pages, pricing pages, scroll motion, WebGL effects, layout systems, and dozens of visual styles.

How PMs use it

When you know the feel you want but cannot describe it, record a competitor's page or a site you admire and run the video-to-superprompt skill. You get a detailed prompt Claude can build from. The landing-page and pricing-page skills are a fast start for launch pages and pricing experiments.

Pros

  • Huge library covering landing pages, motion, and visual styles
  • Reference-to-prompt workflows are unique on this list
  • Plain Markdown skills that work in any agent
  • Each skill is narrow, so you only load what you need

Cons

  • No one-command install yet: you copy the folders you want
  • A third of the library is game development PMs can skip
  • Some workflows assume Codex or Aura Build

Install

Terminal

Clone, then copy the skills you want into ~/.claude/skills/

git clone https://github.com/MengTo/Skills ~/.local/share/mengto-skills
cp -r ~/.local/share/mengto-skills/agent-skills/codex/video-to-superprompt ~/.claude/skills/

Verdict: Cherry-pick, do not install everything. Grab the reference-capture workflow and the landing-page skills and ignore the rest until you need them.

Docs and GitHub
16

Refactoring UI Skills

Claude Code plugin

450+ stars

A polish pass before stakeholders see it.

Show details

Refactoring UI Skills packages the principles from Adam Wathan and Steve Schoger's book Refactoring UI as ten focused skills: visual hierarchy, typography scale, color palette, spacing, button hierarchy, visual clutter, empty states, shadows, contrast, and grouping. A refactor-ui meta-skill runs a full design pass. Works in Claude Code, Codex, and Cursor. (Full disclosure: I built this one. It is not affiliated with Tailwind Labs, and the book is worth buying.)

How PMs use it

Your prototype works and looks fine, but something feels off. Run /refactor-ui on the screen for a full review, or call one skill, like the button hierarchy or empty states skill, when you know where the problem is. Each skill explains the principle behind the fix, so you learn to spot the issue yourself next time.

Pros

  • Proven principles from one of the most-read UI design books
  • Run the full pass or one targeted skill
  • Explains why, not only what to change
  • Free, with Claude Code, Codex, and Cursor manifests

Cons

  • Reviews and refines; it does not set a visual direction
  • Best after you already have a working UI to critique
  • Smaller and newer than the other design packs

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add gnurio/refactoring-ui-plugin
  2. /plugin install refactoring-ui-skills@refactoring-ui-plugin
Claude chat
/refactor-ui

Verdict: The finishing pass. Use Taste Skill or Design and Refine to get a direction, then run this before you show anyone.

Docs and GitHub
17

Emil Kowalski's Skills

Skill pack, not a plugin

49k+ stars

When the motion feels off.

Show details

Emil Kowalski built animation and interface work at Vercel and Linear, and made the Sonner toast library. His skills list the small mistakes agents make and how to fix them, like an ease-in curve on an enter animation or a solid border where a soft shadow belongs. The main emil-design-eng skill covers animation and design, and focused skills build animations, review or improve existing ones, find places that deserve motion, make a web app feel native on a phone, and pick a trusted UI library instead of hand-rolling one.

How PMs use it

Motion is where AI prototypes give themselves away: everything slides in the same way, or nothing moves at all. Run improve-animations on your prototype for a prioritised list of fixes any agent can carry out. Use animation-vocabulary to learn the words that get the animation you picture, so your feedback to Claude, or to a designer, gets specific.

Pros

  • Expertise from shipping interfaces at Vercel and Linear
  • Strict animation review and audit skills for existing code
  • mobile-native fixes the details that make web apps feel cheap on phones
  • pick-ui-library stops the agent installing abandoned packages

Cons

  • Mostly about motion, so less help with layout and hierarchy
  • Some skills are niche, like Swift and React Native
  • Gains are subtle, so they matter most once the basics are right

Install

Terminal
npx skills@latest add emilkowalski/skills

Verdict: The polish layer for motion. Add it once your prototype looks right and you want it to feel right too.

Docs and GitHub
18

Tufte VDQI

Claude Code plugin

~310 stars

When the prototype or deck has charts.

Show details

Two skills built on Edward Tufte's The Visual Display of Quantitative Information. tufte-critique scores a chart against a nine-criterion rubric, names the chartjunk, computes the lie factor and says whether the data wants a different chart type. tufte-chart draws an SVG in Tufte's genres: time series, small multiples, quartile plots and range-frame scatters. Scripts need only Python 3. (Full disclosure: I built this one.)

How PMs use it

Before a metrics chart goes into a deck or a prototype dashboard, ask "is this chart misleading?" and fix what tufte-critique flags, or ask tufte-chart to redraw it.

Pros

  • Critique and rebuild in one plugin
  • Catches charts that mislead, not only ugly ones
  • No dependencies beyond Python 3

Cons

  • Charts only, not layout or UI
  • No license file in the repo yet

Install

Terminal
claude plugin marketplace add gnurio/tufte-vdqi-plugin
claude plugin install tufte-vdqi@tufte-vdqi-marketplace

Verdict: Add it when your prototype or deck has charts people will make decisions from.

Repo

04

Engineering-workflow plugins PMs should know

  • Matt Pocock's Skills: run /grill-me before you build so Claude asks what you meant.
  • Addy Osmani's Agent Skills: a senior engineer's spec-to-ship workflow for when a prototype becomes a real feature.
  • Ponytail: keeps Claude from over-building; run /ponytail-review before you hand a branch to engineers.

Pick · Engineering

19

Matt Pocock's Skills

Claude Code plugin

301k+ stars

Install it and make /grill-me a habit before every non-trivial request. It fixes misalignment at the cheapest possible moment.

Matt Pocock (Total TypeScript, AI Hero) open-sourced the skills he uses daily. They are deliberately small and composable instead of owning your whole process. The most popular are /grill-me, which makes the agent interview you in detail before any work starts, and /grill-with-docs, which does the same while building a shared vocabulary for your domain. A setup skill connects them to your issue tracker (GitHub, Linear, or local files).

How PMs use it

The most common failure with AI agents is that they build something you did not mean. Run /grill-me before any feature, spec, or even a strategy doc, and Claude asks the questions a good engineer would ask in refinement. It works for non-code work too, which makes it the most useful engineering skill for PMs.

Pros

  • /grill-me alone is worth the install, and it works for non-code work
  • Small skills you can read, adapt, and combine
  • In Claude Code's official marketplace, with automatic updates
  • Connects to GitHub, Linear, or local files for triage

Cons

  • Engineering-flavoured: some skills assume you work in a codebase
  • Pick the plugin or the skills.sh install, not both (you get duplicates)
  • Less hand-holding than a full workflow system

Install

Claude Code chat
/plugin install mattpocock-skills
Claude Code chat

Then, once per repo:

/setup-matt-pocock-skills
Terminal

Other agents:

npx skills@latest add mattpocock/skills
Docs and GitHub
20

Addy Osmani's Agent Skills

Claude Code plugin

109k+ stars

A senior engineer’s workflow from spec to ship

Show details

Addy Osmani, an engineering leader at Google, packaged the workflows and quality gates senior engineers use into 25 skills. Nine slash commands map to the development lifecycle: /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify, and /ship. Skills also trigger on their own, so building UI pulls in frontend guidance and designing an API pulls in interface design. An interview-me skill questions you about requirements one question at a time.

How PMs use it

PMs already think in specs. Run /spec on a feature idea and Claude writes the spec before any code, then /plan breaks it into small tasks you can read and challenge. That gives you a checkpoint where your product judgement matters most, before the agent commits to an approach. /ship walks through the checks you would expect before anything goes live.

Pros

  • /spec and /plan put a written spec in front of every build
  • Tests and per-task commits are built into /build
  • Skills trigger automatically for UI, API, and performance work
  • Installs into Claude Code, Cursor, Codex, and many other agents

Cons

  • Overlaps with Matt Pocock’s skills, so pick one as your main workflow
  • The full lifecycle is heavier than a quick prototype needs
  • Marketplace install needs GitHub SSH keys or the HTTPS workaround

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add addyosmani/agent-skills
  2. /plugin install agent-skills@addy-agent-skills
Terminal

Other agents:

npx skills add addyosmani/agent-skills

Verdict: The most complete engineering workflow on this list. Install it when your prototypes start turning into real features someone will maintain.

Docs and GitHub
21

Ponytail

Claude Code plugin

174k+ stars

Makes your agent think like the laziest senior dev in the room

Show details

Ponytail puts a minimalist senior engineer inside your agent. Before writing code it checks whether the browser, the platform, or an installed dependency already solves the problem. Ask for a date picker and you get a native date input instead of a new library and a wrapper component. The author's benchmark on real Claude Code sessions measured about 54% less code and 20% lower cost on average, with safety checks kept. It also ships review, audit, and tech-debt skills.

How PMs use it

Prototypes rot when the agent adds a library for every small feature. With Ponytail on, your prototypes stay small enough to understand, cheaper to run, and easier to hand to an engineer. Run /ponytail-review on a branch Claude built to find what could be deleted.

Pros

  • Less code means fewer places for a prototype to break
  • Measured savings on tokens, cost, and time, with a published benchmark
  • Review, audit, and debt skills for existing code
  • Works across Claude Code, Codex, and many other agents

Cons

  • The headline savings depend on how much your agent over-builds
  • Minimalism can clash with a team that wants explicit abstractions
  • Benchmark ran on one repo and a small model, so treat numbers as directional

Install

Claude Code chat· one prompt each
  1. /plugin marketplace add DietrichGebert/ponytail
  2. /plugin install ponytail@ponytail

Verdict: An easy always-on install. It quietly removes the most common source of bloated AI-built prototypes.

Docs and GitHub

05

Install safely

Skills and plugins are instructions and sometimes scripts that run on your machine.

  • Read the SKILL.md and any scripts before installing.
  • Prefer official sources and active repos with visible maintenance.
  • Be careful with hooks: they run commands automatically on events like saving a file.
  • Don't paste credentials into skill files; use your editor's MCP configuration.

Skills vs plugins vs MCP servers

A skill is a folder with a SKILL.md file: instructions Claude loads when a task matches. A plugin bundles skills, slash commands, MCP servers, hooks and sub-agents into one install. An MCP server connects Claude to an outside tool such as Linear, Notion or Slack. See Anthropic's docs on skills and plugins.

Skills are cheaper than they look. Only each skill's short description loads at the start of a session; the full instructions load when Claude uses it. (Details in the FAQ on token usage below.)

FAQ

What are the best Claude skills for product managers?

Start with Anthropic's free Product Management plugin, then phuryn/pm-skills for breadth. Add Evals Skills if you own an AI feature, and a focused plugin such as Porter strategy skills or PMF Skills when you need one job done well. AI PM OS is the paid option with context and reviewers wired in.

Are Claude Code plugins free?

Most are. Everything on this page is free except AI PM OS ($99 one-time). Claude Code itself needs a paid Claude plan.

What's the difference between a plugin, a skill and an MCP server?

A skill is a reusable instruction set Claude loads when a task matches. An MCP server connects Claude to outside tools and data. A plugin packages skills, commands, MCP servers, hooks and sub-agents into one install.

Is there an official Anthropic plugin for product management?

Yes. The Product Management plugin in Anthropic's knowledge-work-plugins repo has seven commands, including /write-spec, /roadmap-update and /stakeholder-update. It's designed for Cowork and works in Claude Code.

Do Claude skills work in Cowork, Cursor or Codex?

Skills use an open SKILL.md format, so most work across tools. Slash commands are often Claude-specific: phuryn's commands, for example, install in Codex but don't run as slash commands there. Check each repo's install notes.

How many skills or plugins should a PM install?

One PM pack, your own context file, and one design plugin if you prototype. Packs with overlapping commands make it hard to know which one ran.

Are third-party skills safe to install?

Treat them like any code from the internet. Read the SKILL.md and scripts, prefer official or actively maintained repos, and be cautious with hooks that run commands automatically.

How do Claude Code plugins affect token usage?

Skill descriptions load at session start (low cost). Full content loads only when a skill is invoked. MCP server tool names load at start; full schemas load on demand. Hooks run externally and add zero token cost unless they return context. Large skill packs (like MengTo Skills or the full Taste Skill pack) have a slightly higher baseline context load, so install only the skills you use.

Can I build my own Claude Code skill?

Yes. A skill is a folder with a SKILL.md file describing when to use it and what to do. See how to organise Claude Code skills.

Claude skills vs Claude Projects: which should a PM use?

Projects keep files and instructions for chats in the Claude app. Skills are reusable procedures an agent loads when the task matches, and they work in Claude Code and Cowork. Many PMs use Projects for quick chats and skills for repeat work.

Can I build my own Claude Code plugin?

Yes. A plugin is a directory with a .claude-plugin/plugin.json manifest and optional folders for skills, agents, hooks, and MCP servers. You can start by packaging existing CLAUDE.md skills you have already written. The official docs have a quickstart that walks through building a plugin from scratch.

This page is updated as the Claude Code plugin ecosystem evolves. For the official plugin directory, visit claude.com/marketplace/plugins. For MCP setup and connecting Claude to external tools, see the MCP guide for PMs.