Updated September 01, 2026
Claude Code is Anthropic's terminal agent for file-based work. Get started in about 30 minutes: skip the blank CLAUDE.md, run one real PM workflow, then add skills, plugins, and MCP. AI PM OS is the packed version if you want to skip DIY.
Scope: setup, workflows, starter skills, a curated free skills directory, plugins, and MCP. AI PM OS is optional, one checkout sits after the FAQ if you want the full packed system.
The practical guide
Claude Code is Anthropic's terminal-based AI assistant. This guide, updated September 01, 2026, is the getting-started map for product managers: first 30 minutes, skip the blank CLAUDE.md, five file-based PM workflows, plugins, and MCP. Detailed setup lives on the Claude Code setup guide. Curated, not exhaustive.
Set up Claude Code as a product manager in about 30 minutes: install the CLI, open a product folder, skip the blank CLAUDE.md with a starter template, and run one real PM workflow on a file you already have. The step-by-step lives on the setup guide. AI PM OS is the packed version of that layer.
Most ChatGPT answers, and most official walkthroughs, hand you an empty CLAUDE.md and an empty folder tree and call it a PM operating system. That is where people stall. Start from a template, run one real workflow tonight, and add MCP later.
Install Claude Code (under five minutes)
Anthropic's official install guide covers the command and screenshots. Authenticate, type claude, confirm it opens.
Open a product folder, not a blank tree
A folder named product-hub/ is enough. That folder is the context container. Do not spend the first night inventing a 16-folder "PM OS" from scratch.
Start from a template, not an empty file
Paste a starter instead of staring at an empty file. The starter template below below, or the free AGENTS.md template, is the fast path.
Run one PM workflow on a file you already have
Meeting notes or an interview transcript. Ask Claude Code to structure it to a new markdown file. That is the loop you will run daily. Full checkable steps live on the setup guide.
Step by step
The full PM setup guide
Install, CLAUDE.md, MCP, and your first PM workflow โ every step with a checkable outcome.
Full walkthrough
How to Use Claude Code โ the starter guide
Setup, first workflows, and the common pitfalls people hit on day one.
The five recipes cover interview synthesis, competitor monitoring, PRD drafting from Linear tickets, release-note generation, and funnel analysis. Each includes the exact Claude Code prompt and expected saved output. Swap in your own file paths, then run it.
Claude Code prompt
Read all files in /research/interviews/. For each transcript, extract: - Interviewee role and company context - Top 3 pain points with direct quotes - Their current workflow - Feature requests mentioned Then create /research/interview-synthesis.md with: - The 5 most common pain points ranked by frequency - Contradictions or outlier perspectives worth flagging - 3 hypotheses worth testingShips as a slash command in AI PM OS โ
Claude Code prompt
Open each URL in /research/competitors.md. For each competitor, check their pricing page, features page, and recent changelog or blog. Compare against our positioning in /context/product.md. Write a digest to /research/competitor-digest-[today's date].md with: - Any pricing or packaging changes since last check - New features or announcements - Shifts in messaging or positioning - One "watch this" signal worth flagging to the teamShips as a slash command in AI PM OS โ
Claude Code prompt
Using the Linear MCP, pull all open tickets in [project name]. Read /context/product.md and /context/users.md for background. Draft a PRD to /work/prd-[feature-name].md covering: - Problem statement grounded in the ticket patterns - Proposed solution with scope boundary - Success metrics - Open questions for engineering - What is explicitly out of scope Flag any tickets that seem contradictory or under-specified.Ships as a slash command in AI PM OS โ
Claude Code prompt
Use the GitHub CLI to fetch the diff for this commit: [paste commit URL] Write a user-facing release note to /releases/[date]-[feature].md: - Plain-English title (no ticket refs or internal jargon) - 2โ3 sentences: what changed and why it matters to the user - Who this change affects - Any action required from users Match the voice and length of examples in /releases/examples/ if that folder exists.Ships as a slash command in AI PM OS โ
Claude Code prompt
Read @funnel-data.csv. Answer the following and save results to /research/funnel-[today's date].md: - Where is the biggest drop-off by percentage? - Which step loses the most users in absolute count? - What is overall funnel conversion from first to last step? - If we fixed just the worst drop-off step, what would overall conversion look like? Show step-by-step counts and % drop-offs. Add a 3-sentence summary a non-technical stakeholder can read.Ships as a slash command in AI PM OS โ
In AI PM OS: Every recipe above ships as a slash command โ /workflow:discovery, /workflow:competitor, /workflow:prd-draft, /workflow:release-notes, /workflow:funnel. Plus five more covering decisions, stakeholder politics, measurement, document review, and coaching.
CLAUDE.md sits in your project folder and tells Claude Code about your product, tools, and preferences โ permanently. Save this starter into product-hub/CLAUDE.md and fill in your details:
## Product Context Product: [Your product name and one-line description] Stage: [Early / Growth / Scale] Current focus: [Q2 OKR or current sprint theme] PM tools: [Linear / Jira / Notion / Confluence] ## My Workflow Preferences PRDs: Concise. Problem -> solution -> success metrics -> open questions. Research synthesis: Always extract direct quotes alongside pain points. Stakeholder updates: Executive summary first, full detail below. Release notes: User-facing language only, no ticket refs or jargon. ## Key Files - /context/product.md - product overview and strategy - /context/users.md - target personas - /context/team.md - engineering constraints and team norms - /research/ - interview transcripts, survey data, competitive analysis - /templates/ - PRD, release notes, meeting agenda templates - /work/ - active work in progress ## Connected Tools - Linear MCP: [team name, relevant project IDs] - GitHub MCP: [repo name] - Notion MCP: [workspace name] ## Output Defaults Save all outputs to the relevant /work or /research folder. Ask before overwriting an existing file. Default to markdown unless I specify otherwise.
Free template
AGENTS.md for product managers
A complete, paste-ready AGENTS.md plus four sample PM skills. Works with Claude Code and Codex. No email gate.
Sixteen patterns
Best Practices for CLAUDE.md / AGENTS.md
Naming conventions, command shortcuts, and the anti-patterns that quietly poison sessions.
The best free starting points are Anthropic's Product Management plugin, Dean Peters' Product-Manager-Skills repo, Paweล Huryn's MIT pm-skills pack (68 skills, 42 commands, nine plugins), and Carl Vellotti's free Claude Code for Product Managers course. Review plugin permissions before you install anything that ships tools or MCP servers.
Two free starter files, copy or download, then save each as SKILL.md in its own skill folder.
Turn use cases (or a persona + goals) into INVEST stories with Given/When/Then acceptance criteria for happy path and edge cases.
---
name: gherkin-stories
description: >-
Use when you have a use-case list (or a persona, goals, and product context)
and need INVEST-compliant user stories with embedded Given/When/Then
acceptance criteria โ atomic, traceable, covering the edge case as well as
the happy path. Not for pulling a team out of solution mode
(`problem-first`), or for plain role-goal-benefit stories with no scenarios.
---
# Write one atomic story per use case, scenarios embedded
## Step 1 โ Take the use-case list
Get the use-case list (actor, goal, scenario per case), or โ standalone โ the persona, their goals, and product context.
Done when every use case or goal in scope is identified as a story source.
## Step 2 โ Write one atomic story per source
For each source, write a summary and the As-a/I-want/So-that use case, with exactly one trigger and one outcome for the happy path. If a source needs more than one trigger or outcome to describe it, split it into separate stories instead of forcing them together.
Done when every story has exactly one trigger and one outcome, and no story required merging two distinct outcomes.
## Step 3 โ Add Gherkin acceptance criteria
For each story, write the happy-path scenario as Given/When/Then, plus one edge-or-negative scenario (a risk, error, or permission case) also as Given/When/Then.
Done when every story has both a happy-path and an edge/negative Gherkin scenario, each independently testable.
## Step 4 โ Note what QA and engineering need
For each story, note any non-functional requirement (performance, security, accessibility), the event or property to instrument if relevant, dependencies (APIs, teams, flags), and anything explicitly out of scope.
Done when every story's notes are filled in or explicitly marked not applicable โ never silently skipped.
## Step 5 โ Trace each story to its source
Build a table mapping each use case or goal to the story ID(s) that cover it, and how the acceptance criteria prove it.
Done when every source from Step 1 has at least one story, and every story traces to a named source.
## Output
Ask where to save the file (default `./gherkin-stories-{initiative-slug}.md` in the working directory). Write the doc there.
The doc holds: the numbered stories with their Gherkin scenarios and notes, and the traceability table.
Pull a team out of premature solution mode: diagnose the jump, extract the job, challenge assumptions, frame the problem, and name the research that would validate it.
---
name: problem-first
description: >-
Use when a team jumps to solutions without defining the problem โ reorient
them to rigorous problem definition and user job alignment before any design
work. Not for writing Gherkin acceptance stories (`gherkin-stories`) once
the problem is already clear and scoped.
---
# Frame problems before jumping to solutions
Premature solution mode smuggles assumptions. The job is to make those assumptions visible, frame the real problem, and only then reopen the solution space.
## Required inputs
Get the proposed solution and the product/user context. Ask for either if missing.
Done when both are on the table in the team's own words.
## Step 1 โ Diagnose the solution jump
Name the proposed solution, the assumptions embedded in it, the implied problem it is trying to solve, and whether that problem has been validated with users.
Done when each of those four is written explicitly โ including "not validated" when that is the truth.
## Step 2 โ Extract the underlying problem
From the jump, pull out the user need, the job-to-be-done, the desired outcome, how users do this today (and what goes wrong), and the real constraints.
Done when need, job, outcome, current state, and constraints are each stated without naming a solution.
## Step 3 โ Challenge the assumptions
For every key assumption from Step 1: is it validated (yes / no / partially), what evidence exists, what happens if it is wrong, and the cheapest way to test it.
Done when every key assumption has those four fields filled.
## Step 4 โ Frame the problem statement
Fill: Who experiences it ยท What friction ยท When/where ยท Impact ยท Current workarounds ยท What success looks like. Then write one sentence: "Users [who] struggle to [what] when [context] because [root cause], which leads to [impact]. Success would mean [outcome]."
Done when every field has a real value or an explicit TBD, and the one-sentence statement uses no solution language.
## Step 5 โ Offer alternative framings
Write two or three alternative ways to frame the same situation that would open different solution spaces.
Done when each framing would clearly lead a team toward a different class of solution.
## Step 6 โ Name the validation research
List 5โ7 research questions that would confirm or kill this problem, a method (interviews, observation, diary, analytics), success criteria for "this is the right problem," and a rough timeline. Frame the redirect as de-risking the solution, not rejecting ideas.
Done when every question maps to an assumption or a TBD, and the method names who to talk to or what to measure.
## Output
Ask where to save the file (default `./problem-first-{topic-slug}.md` in the working directory). Write the doc there.
The doc holds: solution-jump diagnosis, underlying problem, assumption challenges, problem statement, alternative framings, validation research, and a short draft message the PM can send the team to redirect the conversation.
Want a deeper prodmgmt.world sample (owned by George)? See make-requirements-great.
Last checked: 2026-08-04. Comparison criteria: best for, scope, delivery/install shape, and license or usage status.
Plugins can include tools or MCP servers. Review plugin permissions before installing anything from a third-party repository.
Verified on 2026-08-04: public product-management plugin path under anthropics/knowledge-work-plugins
Verified on 2026-08-04: public deanpeters/Product-Manager-Skills repository
Verified on 2026-08-04: public phuryn/pm-skills README scope claim
Verified on 2026-08-04: ccforpms.com course + carlvellotti/free-ai-courses source
Source repo: free-ai-courses
Organize them
How to organize Claude Code skills as a PM
The .claude/skills/ folder structure, grouping by PM workflow, and why routing, not skill count, sets the ceiling on a collection.
The plugins worth installing first, organized by what you'll use them for, not by who built them.
| Plugin | Use Case |
|---|---|
| Product Management | |
| PM Skills Library โ | Reusable skills for strategy, research synthesis, communication, and execution |
| Frontend Slides โ | Create beautiful slides on the web using Claude's frontend skills |
| NotebookLM MCP CLI โ | MCP server integration for Google NotebookLM - connect your AI workspace to Claude Code |
| Evals Skills โ | Skills for AI evals โ companion to the "AI Evals For Engineers & PMs" course; covers setting up rigorous quality checks for agent workflows |
| Engineering | |
| Superpowers โ | Curated Claude Code plugin marketplace โ TDD, debugging, collaboration patterns, and proven engineering techniques |
| Compound Engineering Plugin โ | Official plugin for engineering workflow automation โ works with Claude Code and more. Covers planning, review, and knowledge capture |
| Get Shit Done โ | Meta-prompting system that transforms vague ideas into shipped features through spec-driven workflows, multi-agent orchestration, and atomic git commits |
| Vercel Agent Browser โ | Browser automation CLI for AI agents |
| Vercel Agent Skills โ | Official collection of agent skills from Vercel Labs |
| Shaping Skills โ | Claude Code skills for Shape Up methodology - iterate on problems and solutions before implementation with /shaping and /breadboarding workflows |
| gstack โ | Garry Tan's exact Claude Code setup โ 23 opinionated tools acting as CEO, Designer, Eng Manager, Release Manager, Doc Engineer, and QA |
| Design | |
| Design Plugin โ | UI design and prototyping with Claude Code |
| Layers โ | Product design skills for making decisions beneath the screen โ strategy, user needs, conceptual models, flows, and domain language |
| Emil Kowalski Skill โ | Design engineering guidance for better interfaces, animations, performance, and frontend craft inside Claude Code and other agents |
| User Interface Wiki โ | Living manual for better interfaces โ curated guidelines for building and reviewing UI with AI agents |
| UI Design Brain โ | Best practices, layout patterns, and design-system conventions for 60+ interface components โ production-grade UI instead of generic output |
| ui-skills.com โ | Opinionated constraints for building better interfaces with agents |
| SuperDesign โ | Design discovery and generation - search design prompts, extract brand guidelines, and create design variations on an infinite canvas |
| Impeccable โ | Premium frontend design skill with commands for typography, color, motion, polish, audits, browser iteration, and anti-pattern detection |
| Misc | |
| Skills.sh โ | Open agent skills ecosystem โ one-command installation, compatible with 20+ agent platforms, community-maintained skill leaderboard |
Deep dive
The complete plugins guide
Reviews, install commands, and PM-specific use cases for every plugin in the table above.
In AI PM OS: Plugins are individual tools you install one at a time. AI PM OS is the operating system that orchestrates them โ 243 skills indexed by context, invoked on demand.
Model Context Protocol (MCP) servers let Claude Code reach into Linear, GitHub, Slack, Notion, and beyond. Query tickets, pull diffs, summarize threads, reference docs, from inside one prompt.
A real example: โPull all checkout-related Linear tickets from the last 30 days, show the payment-flow implementation from GitHub, summarize the #checkout-project Slack channel, and draft a PRD for the next iteration.โ One prompt, three MCPs, one PRD on disk.
Directory
MCP servers, curated for PM work
Linear, GitHub, Slack, Notion, Figma, with setup notes for each.
Do not confuse the two
Linear MCP is not an AI PM OS
Linear MCP is where work lives. An AI PM OS is how PM thinking runs. Fair 2026 comparison.
Cluster links: setup, AGENTS.md, skills, the Cursor comparison, AI PM OS vs Linear MCP, and Anthropic's canonical docs.
Starter Guide: How to Use Claude Code โ
Setup walkthrough, first workflows, and common pitfalls, the on-ramp before you wire anything custom.
AI PM OS vs Linear MCP โ
Linear MCP connects where work lives. An AI PM OS encodes how PM thinking runs. Fair 2026 comparison, with a clear winner.
AI for Product Management: First 5 Workflows โ
The cluster hub, where Claude Code fits in the five starter AI PM workflows before picking a full tool stack.
Claude Code vs Cursor for Product Managers โ
Head-to-head on pricing, context windows, usage limits, and which tool fits PM work. Most AI-forward PMs run both.
15 AI Tools Product Managers Use Daily โ
Where Claude Code sits in the full 2026-tested AI stack for PMs, with pricing, free tiers, and verdicts.
I Built 3 Claude Code Commands to Speed Up My PM Work โ
Custom commands for the PM-specific rituals, standup notes, PRD scaffolds, weekly digest.
Workflows, Not Skills: How to Chain Claude Code Commands โ
Why skills alone plateau and how chained workflows compound across a PM week.
Claude Code in Action (Anthropic Academy) โ
Free course: architecture, context management, custom commands, MCP servers, GitHub integration, hooks.
Best Practices for Claude Code (Official) โ
Anthropic's canonical guide, environment setup, context management, verification, planning, communication patterns.
Common Workflows (Official) โ
Step-by-step guides for exploring codebases, fixing bugs, refactoring, testing, PRs, Plan Mode.
Plugins Reference (Official) โ
Technical reference for the plugin system: skills, agents, hooks, MCP servers, manifest schema.
Claude Code setup and workflow questions first, then buying questions about AI PM OS.
Claude Code is Anthropic's AI coding assistant that runs in your terminal. Product managers use it to automate research synthesis, draft PRDs, watch competitors, write release notes, and build light internal tools. You do not need to write code yourself. It understands your full project context and acts on it.
No. Most PM workflows in Claude Code use plain English instructions. You describe what you want, for example 'read these interview transcripts and find the top five pain points.' Then Claude Code executes it. Some setup (installing the CLI, connecting MCP servers) is straightforward, and skills and plugins handle the rest.
Claude Code works directly on your files and tools. It can read a folder of interview transcripts. It can write a summary to a new file. It can pull data from a Linear board via MCP. It can chain those steps together on its own. ChatGPT and Claude.ai are conversational. Claude Code is agentic.
The best free starting points are Anthropic's Product Management plugin, Dean Peters' Product-Manager-Skills repo, Paweล Huryn's MIT pm-skills pack (68 skills, 42 commands, nine plugins), and Carl Vellotti's free Claude Code for Product Managers course. Review plugin permissions before you install anything that ships tools or MCP servers. Free starter examples on this page are gherkin-stories and problem-first โ the same craft as AI PM OS skills, adapted to run standalone. AI PM OS packages 243 PM skills when you want the full operating system.
Common PM automations include: synthesizing user interview transcripts, drafting PRDs from Linear tickets, monitoring competitor pricing pages weekly, generating user-facing release notes from commits, analyzing funnel data from Mixpanel or Amplitude, and building lightweight internal dashboards. AI PM OS packages 243 PM skills, 150+ frameworks, and 11 guided workflows โ all triggered by your context, so you spend zero time hunting for the right prompt.
AI PM OS is a product management operating system that runs inside Claude Code. After onboarding, you create curated context about your company, product, team, goals, and constraints. From then on, each response starts with your situation instead of a blank prompt.
The workspace includes 243 PM skills, 150+ frameworks, 11 guided workflows, review subagents, and live connectors โ one complete operating system for one PM.
No. Linear MCP connects Claude Code or Claude Code to your Linear workspace so the agent can find, create, and update issues. That is where work lives. An AI PM OS encodes how PM thinking runs: skills, workflows, frameworks, and review subagents.
They complement each other. If the team already runs on Linear, connect Linear MCP and run AI PM OS for craft. The full comparison: AI PM OS vs Linear MCP.
A skill is a single AI-invocable PM task โ for example, /skill:jtbd-interview runs a structured Jobs-to-be-Done discovery session.
A framework is a structured thinking model a skill applies โ for example, one of the 50 prioritization frameworks, each tagged with 'Use When' metadata so the right one activates for the situation.
A workflow is a chained, multi-step process โ for example, /workflow:strategy sequences discovery โ opportunity โ PRD as a connected session, not three separate prompts.
PM OS ships with 243 skills, 150+ frameworks, and 11 workflows โ organized so the right one surfaces when a PM needs it.
AI PM OS is a packaged workspace of files and workflows your editor reads automatically. No manual hunting through prompt libraries.
243 PM Skills โ across 12 categories including strategy, discovery, prioritization, stakeholder management, and metrics.
150+ Frameworks โ organized by product stage: discovery, validation, build, and grow.
50 Prioritization Frameworks โ each tagged with "Use When" metadata.
100 Customer Interview Questions โ across 6 research categories.
41 North Star Metric Examples โ from real growth-stage companies.
25 PM Practice Drills โ with session tracking and progress logs.
7 PRD Templates โ Amazon PRFAQ, Shape Up pitch, Feature PRD, and more.
11 Guided Workflows โ end-to-end motions you can run the same way every time.
260 Curated Lenny Posts โ surfaced when you are stuck.
5-File Context System โ tell the agents your company, product, team, goals, and constraints once.
Free packs are a real option, and I link to good ones myself. The difference is what you are buying. A free pack is a folder of skill files. AI PM OS is a curated, maintained operating system.
With free packs, the assembly is on you: decide which skills are worth keeping, structure the folders, write your context files, wire up the workflows and MCP servers, then keep all of it current as Claude Code and Claude Code change. That curation and upkeep is the real cost, and it is the part most PMs never finish.
AI PM OS is that work already done: 243 skills curated into 12 categories, 11 guided workflows, 12 review subagents, 6 MCP server configurations, a 5-file context system, and free lifetime updates maintained against current Claude, Claude Code and Claude Code releases. The $99 buys you out of months of assembly, not the files themselves.
Want to check the quality bar first? The free public sample is full skill source, the same standard as the paid workspace.
No. AI PM OS is a folder of Markdown files your editors read. It does not replace your company Claude workspace, wipe custom instructions, or force a single vendor lock-in. Open it alongside whatever stack you already run.
Yes. Checkout runs through Polar, which issues receipts and invoices for business purchases. If your finance team needs a custom invoice or PO trail, email me at george@prodmgmt.world after checkout and I'll help.
No. Claude Code is an AI tool โ you type plain English. Setup is download the workspace from Polar, open it in Claude Code, run /start, and answer a few guided questions about your product. No code, no installs beyond the editor.
Claude Code requires a paid Anthropic plan (or Claude.ai Max subscription). See current pricing: claude.ai/pricing
AI PM OS is a separate license ($99 one-time for one PM). Your editor is a separate product with its own pricing. That's the honest picture before you decide.
AI PM OS is a local workspace of files on your machine โ nothing leaves your computer through me. When you ask Claude Code a question, context is sent to that editor's configured LLM provider under that vendor's terms โ the same as any other prompt. Nothing routes through prodmgmt.world. prodmgmt.world is never in the data path and has no subprocessors for this product. If your company restricts AI tool use, AI PM OS inherits those same restrictions.
Need a procurement packet or DPA conversation? Email george@prodmgmt.world.
For a limited window, Individual includes a setup call with me โ I get the workspace open in your editor and the skills loading. You'll get the booking link after checkout.
Outside that window, email me at george@prodmgmt.world and I'll help you get unstuck.
Onboarding is light: download the workspace from Polar, open it in Claude Code and run /start. The five context files (company, product, team, goals, constraints) are built with guided prompts โ not a blank-page homework assignment.
Yes. AI PM OS comes with a 30-day money-back guarantee. Try it in your real workflow, run it against your actual PM work, and if it doesn't help, email me within 30 days of purchase for a full refund.
No hoops, no hard feelings.
After purchase you download the workspace from Polar, open it in Claude Code and run /start. When updates ship, re-download from Polar.
Skills and workflows are maintained against current Claude Code releases.
Yes. A free public sample of the same quality bar lives at github.com/gnurio/nurijanian-skills. Start with make-requirements-great โ the full skill source, not a teaser. The full AI PM OS workspace is built to that same bar.
$99 one-time for one PM. Free lifetime updates.