PMM OS — Product Marketing Operating System

Turn a product brief into positioning, messaging, personas, pricing, competitive intel, and a full launch plan — then package the whole thing into a shippable, interactive launch kit. Without leaving your terminal.

A Claude Code / Codex plugin: 45 product-marketing, GTM & PLG skills, two built-in research engines, a deep framework library, and a zero-dependency generator that turns strategy into an app.

Why it exists

Most "marketing AI" gives you a wall of generic advice. PMM OS gives you decisions and artifacts: it grounds every output in your product context, applies real frameworks (April Dunford positioning, JTBD personas, value-metric pricing, win/loss CI), holds a strict depth bar so nothing comes back as a one-line stub, and ends a launch with something you can actually hand to a team.

It behaves like one product, not 45 band-aided prompts: shared context, one depth standard, one framework library, and a chain that always flows to a finished deliverable.

The launch kit — what you actually ship

Every launch ends with a single self-contained, offline HTML app (no dependencies, no build step to view it): a left sidebar, a main workspace, and a right inspector — plus a ⌘K command palette, present/deck mode, status tracking, and a one-click live editor (rich text, slash-insert, drag-reorder, highlight→comment) that writes changes back to source.

⌘K command palette — jump to any section, persona, pillar, or battlecard

| Positioning | Competitive | Pricing | |:--:|:--:|:--:| | | | |

The screenshots above are a real, end-to-end launch PMM OS produced for a fictional company (Plotline — self-serve product analytics). The whole thing — positioning statement, message house, battlecards, pricing logic, GTM plan, PLG loops, coach review — lives in one kit-content.json and builds to one HTML file. Open demo/plotline-launch/ to click through it. Not a SaaS? demo/enamel-it/ is the same OS run for a geo-bound B2B service (a dental-IT MSP): local events + channels desks, Google-review VoC, retainer pricing, and a referral engine instead of PLG.

Quick start

Claude Code — one command, which registers the marketplace and installs the plugin for you:

npx pmm-os install

Or do it by hand inside Claude Code (both steps — the first only adds the catalog):

/plugin marketplace add buildingwithai/pmm-os
/plugin install pmm-os@pmm-os

Codex:

codex plugin marketplace add buildingwithai/pmm-os
codex plugin add pmm-os@pmm-os

Then restart, or run /reload-plugins. Plugin skills are namespaced, so they're /pmm-os:<skill>:

/pmm-os:product-marketing-os      run the whole chain: grounding → research → strategy → launch kit
/pmm-os:pmm-positioning-exercise  or jump straight to a specialist
/pmm-os:pmm-pricing-packaging
/pmm-os:pmm-launch-kit

You don't have to use slash commands at all — just say "take Plotline to market" and the right skills get picked up. Ask for a launch and PMM OS produces the strategy and the clickable kit; markdown alone is never the finish line.

npx pmm-os doctor checks your environment · npx pmm-os update upgrades · npx pmm-os setup installs the optional research stack.

Needs Claude Code 2.1.143+ (May 2026) or Codex with plugin support. On Codex, plugin hooks stay inactive until you review and trust them — Codex will prompt you.

The quality architecture (v3) — the consulting-grade bar

v3 wires a consulting-grade operating discipline through the whole pipeline, distilled from the MBB working canon and field-researched one-pager formats:

• The two-product contract — every engagement ships the research library (cited, two-altitude desks; every claim clicks through to its source) and the deliverable suite (positioning one-pager, messaging house, persona cards, battlecards with VARS objection handling, pricing, GTM-on-a-page — each backed by a working doc whose claims @-link to the findings that justify them).
• Stage 0 product grounding — when your product has a codebase, the code is read first; capability claims cite path:line. Truth order: code > founder artifacts > description.
• The Plan gate — no research call before a day-1 hypothesis (with named kill conditions) and a MECE issue tree exist. Pivots are logged, never silent.
• The adversarial pass — every desk hunts for evidence against its own finding before it closes. Refuting the hypothesis is a valid, valuable outcome.
• The score gate — deliverables ship at ≥24/33 on an 11-row scorecard (answer-first, action titles, so-what density, typed + dated evidence, honesty…), coach-reviewed, weakest row first.
• The workspace — the launch kit is a Notion-shaped research workspace: filterable Findings / Evidence / Events databases, a side peek that opens any citation's source record, @-mentions with backlinks, markdown shortcuts, and an in-kit Generate with AI queue that a Claude Code session fulfills against the evidence.

What's inside

45 skills, organized as one system:

| Area | Skills | |---|---| | Research (built-in) | pmm-research-brief · pmm-research-desk · last30days · agent-reach | | Orchestrators | product-marketing-os · agentic-marketing-orchestrator · product-lifecycle-os · plg-gtm-strategy | | Positioning & messaging | pmm-positioning-exercise · pmm-positioning-audit · pmm-messaging-positioning · pmm-messaging-hierarchy · pmm-message-market-fit · pmm-adaptive-messaging | | Research & audience | pmm-customer-research · pmm-personas · pmm-icp-definition · pmm-voc-synthesis · pmm-product-context | | Competitive | pmm-competitive-intelligence · pmm-battlecard · pmm-competitive-landing-page | | Pricing | pmm-pricing-packaging · pmm-pricing-analysis | | Launch & GTM | pmm-go-to-market · pmm-launch-brief · pmm-campaign-brief · pmm-feature-announcement · post-launch-learning-loop | | Sales & GTM engineering | sales-enablement · pmm-sales-narrative · pmm-outreach · gtm-account-research · gtm-icp-scoring · gtm-signal-campaign · gtm-weekly-context-update | | Content & SEO/AEO | pmm-content-writer · pmm-aeo-geo · osp-value-map · osp-content-optimizer · osp-technical-marketing | | Production | pmm-artifact-factory · pmm-launch-kit · prd-prototype-factory · pmm-coach |

Three things make them work as a unit:

• A deep framework library — ~76k lines of frameworks, worksheets, templates, checklists and glossaries across 9 domains (positioning, messaging, competitive intelligence, personas, pricing, product launch, rollout, sales enablement, strategic thinking). 43/45 skills cite the exact docs they apply.
• One depth standard — every section must be specific, complete, reasoned, and evidenced — never a generic one-liner. Enforced in the skills, the prompt hook, and a final QA gate. "Depth is not length" — it won't pad, it goes deeper.
• Smart routing — sharpened, disambiguated triggers ("use this, not that") plus a hook that surfaces the right skills and always routes a launch to the interactive kit.

Built-in research (two engines, on purpose)

Good marketing starts with what real people actually say. PMM OS ships two independent, complementary research engines — run them separately and you get two source pools instead of one:

• last30days — an engagement-ranked, last-30-days engine. Pulls Reddit, X, YouTube, TikTok, Hacker News, Polymarket, and GitHub, scores by upvotes, likes, and real money, and synthesizes a brief with verbatim community quotes. Best for "what are people actually saying / sentiment / what's trending." Self-contained Python (3.12+); Reddit/HN/Polymarket/GitHub work zero-config.
• agent-reach — a breadth-first fetch/read router across 13 platforms, strong exactly where the other is thin: GitHub/dev, video transcripts, reading any URL, and Chinese platforms (XiaoHongShu, Bilibili, Xueqiu, V2EX, Xiaoyuzhou). Best for "fetch/read this, search this platform." Opt-in: npx pmm-os setup (it installs Python packages and a headless browser, so it asks first).

Both are vendored into the plugin so research works on a fresh marketplace install with no quality degradation — same depth as running them standalone. Refresh from upstream anytime with bash scripts/sync-research-engines.sh. For VoC, competitive, or any "what do people think" task, PMM OS runs both and synthesizes.

Research Desks — research that produces artifacts

It starts with a brain-dump. You rarely have clean parameters — you have "I've got a product called X, it does A/B/C, I want to launch it." pmm-research-brief is the front door: it distills that ramble into a Product Brief + a complete, mandatory Research Plan across all ten desks (product, customer, competitive, market, pricing, channels, analyst/KOL, events, reviews, GTM/launch) — scoped with named competitors, segment, and events — then runs the full sweep. Nothing is optional; a thorough PMM researches all of it. From there:

Research isn't one undifferentiated blob. A research desk (pmm-research-desk) is a domain owned by a senior specialist's head. It loads that specialist's question set, fans it out across both engines (~15–30 scoped calls — neither engine takes a batch, so the desk decomposes), captures sourced evidence into .agents/research/evidence.md, and emits a structured artifact into the dashboard. The Events Desk, for example, ranks the conferences your ICP actually attends, by relevance vs. cost:

That artifact is built from a template the skills already carry (22 fill-in templates — battlecards, canvases, pricing matrices, RACI… — plus 13 one-pager formats), hydrated by the evidence.

All ten desks ship — product, customer, competitive, market, pricing, channels, analyst/influencer, events, reviews, and GTM/launch — each producing its skill's real deliverable in the dashboard (a competitor matrix + battlecard, a persona one-pager + ICP, a pricing comparables matrix, a SWOT / Five-Forces / PESTEL category brief, a ranked channel map, a KOL list, a ranked events table, a launch-tactics map…):

And research is the hard gate before strategy: a launch / positioning / pricing deliverable won't ship until the desks that feed it have run and the ledger holds their sourced sections — so output is grounded and cited, never invented.

How a launch flows

brain-dump → research brief → research desks → positioning → messaging → pricing → competitive
      → GTM plan → artifacts → PMM Coach review → ✅ interactive launch kit → measurement → iteration

A launch or GTM motion always ends by building the kit (the wiring makes it the required final step), so you finish with a clickable hub, not a folder of docs.

Build a kit from your own strategy

The generator is zero-dependency Node. Author one kit-content.json and build — no copying:

node skills/pmm-launch-kit/scripts/build-kit.mjs <your-launch-folder>
# → <wordmark>-launch-kit.html  +  generated-docs/*.md  +  deck.md (Marp → pptx/pdf)

Everything is one source of truth: kit-content.json → HTML app + markdown mirrors + a slide deck, all driven by a uniform block registry, so new block types and new export targets are additive.

Credits & license

MIT © Jovanny Tovar. PMM OS adapts and unifies a number of open marketing skill sets — most notably the deep PMM framework library from buildingwithai/skills (MIT) — and includes methodology from Open Strategy Partners (CC BY-SA 4.0). All upstream attributions are preserved in THIRD_PARTY_NOTICES.md. See LICENSE.