marketing-mindset

Marketing Mindset

The marketing OS for AI agents — think like a marketer first, get tactics as the output. Not another bag of CRO/SEO/copywriting tricks. This is how a 15-year B2B marketer decides, so an agent gives a real opinion instead of a template.

📄 Landing page & docs: https://axelfreeman.github.io/marketing-mindset/?utm_source=github&utm_medium=readme&utm_campaign=mindset

Install in one command: npx skills add axelfreeman/marketing-mindset

---

---

Test limits — how much volume before a test can be judged

Below the limit you are measuring randomness, not the market. Declare the volume before the test starts.

The rule that follows: what comes easy, scale it. The channel, message or offer that performed noticeably easier than the rest goes first — effort first, money later. You can only see that gap above the limit: below it, "easy" and "ordinary" look identical.

Declare the volume → run one variable → stop at the limit → scale what came easy. Anything else is interpretation.

Full write-up with sources: https://axelfreeman.github.io/marketing-mindset/?utm_source=github&utm_medium=readme&utm_campaign=mindset

---

The gap this fills

Every other "marketing skill" for AI agents hands over tactics. Nobody packages how a marketer thinks. When an agent needs to decide should I do X to get Y, where do I get my first customer, or is this idea worth it, tactics don't answer. This skill does.

See it work

📄 View the demo transcript — one request in ("SaaS for finance, 0 customers"), a sharp 3-month plan out.

The problem

Founders and solo operators can generate code, not content. "How do I do marketing" is genuinely unclear to them. They need an agent that gives honest, non-generic marketing judgment — not a template.

The fix

Six principles and a decision framework, distilled from 15 years of hands-on B2B internet marketing:

1. Don't learn marketing from stale sources — skip the first Google results and cached LLM doctrine
2. Three-month horizon — no 2-year cycles; be useful within 3 months
3. The user has the right to make the first move — don't block bold or hacky first steps
4. Every hypothesis must be testable fast — any teammate can run the test
5. Marketing runs ahead of the product — ship the landing page before the build
6. Marketing never works for free — marketing is exchange; every action must trade for something

Plus: competitors as the source of truth, three keys to the human, and the client stages (first client by hand and free → 2–10 by copying competitors).

The metaphors (share these)

• 🍔 The McDonald's Burger — photograph the product better than it is
• 🧬 Think like a cancer cell — when nothing else applies, multiply
• 🚫 The stop-list — why Product Hunt is lying to you
• 🌑 The Despair Dividend — when every reasonable move fails, the strange hypotheses are the good ones

Who it's for

• AI agents (Claude Code, Cursor, ChatGPT — any agent that reads SKILL.md)
• Founders & solo operators selling a service that can't be touched but is needed right now
• Indie hackers & solopreneurs shipping a SaaS alone
• Freelancers (dev, design, writing) hunting for client #1
• Developers who can ship code but freeze on "how do I get customers"
• Agency owners & fractional CMOs — encode your judgment so juniors stop producing generic work
• Course creators & info-product sellers — the "can't be touched, needed now" market
• Product managers validating an idea before writing code
• Prompt engineers studying how to give an AI a personality with hard boundaries
• VC, investors & venture scouts — stress-test marketing claims in due diligence
• Open-source maintainers — grow adoption of a project
• Startup accelerators & incubators — a repeatable framework for portfolio companies
• SDRs & sales engineers — own their own outreach and positioning

Tested on

Verified to load correctly on these models before release — each adapter matches the target's native syntax:

What you actually get

Working through this skill ends in concrete deliverables — a competitor analysis, three outbound angles, and sharper positioning (see the demo). The mindset is the input; the tactics are the output.

Install

One skill, four harnesses — pick whichever you run:

# skills.sh / npx
npx skills add axelfreeman/marketing-mindset

# DeepSeek Harness (dsh)
dsh plugin --profile <name> add github:axelfreeman/marketing-mindset

Claude Code and Hermes read the same SKILL.md — copy it into your skills directory (~/.claude/skills/marketing-mindset/ or your Hermes skills dir).

Low-context or weaker models? Use the compact SKILL.lite.md — the same mindset compressed to the essentials.

The first-client gate

python scripts/first-client-gate.py

One honest question before you start: do you have your first client yet? If the answer isn't "me," the skill waits.

The playbook — the second step, not the first

Most people ask for a marketing playbook at the worst possible moment: before anything works. Compiling one does not move you closer to sales.

It is the second step — what you write after a hypothesis is already tested and working, when the job changes from finding a channel to draining it: drinking that hypothesis to the bottom, taking the market capacity it holds. The order is fixed: find a working hypothesis by hand → press it to its ceiling → only then write it down.

A playbook is a written set of client-acquisition moves that already work — simple, primitive, repeatable:

"Bought an ad from a blogger in segment A — it produced clients." "Segment B gave the numbers in three months." "This targeting converts a client at break-even or better."

One proven move, one line.

Once those moves are written down they stop being founder-only knowledge. They can be delegated — to a marketing hire, to a department, or to an agent if the work is genuinely repeatable — and that handoff ends free search and starts regular management: goals, volume, cadence, reporting. Critical mass is around six working hypotheses; that is the point at which you can hire marketing.

Two rules the skill will hold you to:

• The playbook is filled from your own experiments only — run by you on your own base. Borrowed case studies and a competitor's playbook are a wish list, not a playbook.
• Failed hypotheses go to the archive, one line each. No reflection, no museum of failures. Marketing is not a mathematical craft yet, so a zero does not tell you which variable was wrong — and handing a new person your failures is toxic, because a marketer starts from zero, not from the previous person's pause point.

Ask for a playbook before a single working hypothesis exists, and the skill will tell you plainly that the job is still the search.

Spread the word

If it works, star it and share it. The network effect compounds.

More from the framework

• competitor-xray — competitor research that ends in a copy-list, not a report
• inbox-audit — will your email reach the inbox? Free DNS audit + paste-ready fixes
• agent-stack-starter — self-hosted AI marketing stack, one job live on day one

Stay updated

Axel Freeman ships releases and improvements to this repo regularly. If this clicks, watch the repo (and star it) — that's how you get each new release and update automatically.

The roadmap is long — active development is planned for at least the next 6 months, so there's a steady stream of updates coming.

Getting first customers is hard — and most marketing advice is generic. Get weekly non-generic teardowns:

• 📣 Telegram: @axelfreeman
• 🐙 GitHub: @axelfreeman

Keywords

marketing mindset, marketing operator mindset, how a marketer thinks, marketing skills for AI agents, growth marketing, first customers, B2B lead generation, positioning, launch, copywriting, cold outreach, SaaS marketing

A note on "growth hacks" and "embellishing"

"Growth hacks", "embellish", and "frame case studies" mean marketing positioning and copywriting — presenting the product's value in its best, boldest light — not fraud, fabricated testimonials, or deceiving users. Bold marketing, always legal.

Author

Written by Axel Freeman, a B2B marketer with 15 years of validated experience across B2B SaaS. Not a content farm — the operator's own method.

License

MIT

Where else this lives

Everything below is public and checkable — pages, packages, articles:

The skill is free and MIT. The paid work is the engineering around it: distribution, measurement, and the uncomfortable part — deciding what has enough volume to be judged at all.