A Claude Code Skill for Positioning: The One That Argues Back

Ask any capable model to help with your positioning and it will help. That is the problem. You describe the product, it tells you the idea is strong, and twenty minutes later you are holding a sentence like "for busy professionals who want to save time." Nothing in that sentence is false. Nothing in it is usable either — you cannot write an ad from it, you cannot pick a subreddit from it, and no stranger reading it can tell whether they are the customer.

The failure is not that the model is weak at positioning. It is that positioning is an interrogation, and a bare chat prompt has no reason to interrogate you. It has every reason to agree with you.

A positioning skill for Claude Code is a fixed process the agent follows instead of improvising a friendly one. This page is about the one AgentSource ships for that job — the Positioning Market Map, $9 — and, as with every skill on the shelf, its captured field report is readable before you pay.

What the skill actually changes

It is a plain-markdown SKILL.md workflow. It writes no code and scaffolds nothing. It produces exactly one file, docs/positioning.md, and it gets there through eight phases the agent runs one question at a time.

The one-question-at-a-time rule sounds like a small stylistic choice. It is the load-bearing one. A model asked six positioning questions at once will answer all six itself, plausibly, and hand you a finished document you never actually thought about.

The gate that runs before anything else

The skill refuses to position a product whose demand is unproven, unless you explicitly override it.

This is worth dwelling on, because it is the opposite of what a sales-minded tool would do. Sharp positioning on an unwanted product does not produce a business; it produces a very well-worded way to describe something nobody asked for. The skill wants evidence first — customer quotes, reviews, revenue, validation notes — and says so before it will start. If you have not done that work yet, the honest sequence is demand first, positioning second.

Four competitor columns, and the one everyone omits

Most competitor analysis is a list of rival tools. The skill requires four columns:

Column	What it captures
Status quo	The workaround in use today — a spreadsheet and email, a manual process, an agency
Direct	Tools solving the same problem the same way
Adjacent	Neighbouring tools that could expand into yours
Outsider	Another industry's solution, an open-source project, a services player

The first column is the one that gets skipped and the one that usually wins. Most products do not lose to a rival; they lose to nothing changes. A matrix that omits the status quo will tell you that you are cheaper and faster than three competitors while the actual buyer keeps using their spreadsheet.

Each entry has to carry a name, a one-liner, a pricing anchor, what they get right, what they get wrong, and where they are vulnerable — and then one sentence naming the vacuum all four columns leave open. "What they get right" is a required field on purpose; a matrix where every competitor is bad is a matrix nobody researched.

The niche drill, graded

The skill defines a niche as the intersection of vertical × use case × stage, and grades answers against that. Its own examples, verbatim:

• ❌ "Marketing teams" — not a niche
• ❌ "B2B SaaS marketing teams" — still too broad
• ✅ "Series-B B2B SaaS marketing teams running outbound on Apollo with <5 SDRs" — a niche

If your answer looks like the first two, the agent keeps asking. Most people stop at the second one and believe they have narrowed, because it is visibly narrower than the first.

A UVP it will not let you fudge

The final statement comes out in Geoff Moore's form, with six slots to fill:

For [the named ICP], who [pain, from evidence], we are [category], that [unique mechanism], unlike [an entry from the matrix], we [the wedge — falsifiable and measurable].

Three rules are enforced around it, and each closes a specific escape hatch:

• "AI-powered" is rejected as a mechanism. It describes your stack, not what the product does differently.
• The "unlike" must be an entry that exists in your matrix. No invented strawman competitor who is conveniently worse than you.
• The agent must say whether the result is sharp or fuzzy, and why. The phrase "nice positioning" is banned output.

That last rule is the anti-sycophancy clause, and it is the part a bare prompt structurally cannot give you. A model that is free to praise your UVP will praise it.

What you actually get out

One draft-stamped file: the upstream read, an ICP one-pager written as a named human with a day shape and a buying committee rather than a category, the four-column matrix with the vacuum named, the business model with a pricing anchor and the path to dollar #1, the niche sentence with a defensibility note, the UVP, and the open questions it could not close. Then a revise loop.

The open-questions section matters more than it looks. A positioning doc with no open questions is usually a doc that guessed.

Where this sits against the alternatives, honestly

• A free prompt or a directory download. Genuinely fine for a first pass, and free. What you tend not to get is the enforcement — the gates, the rejected answers, the banned phrases. Free skill files are also unverified by default, which is why the sensible habit is to read a skill's instructions before you install it, whoever it came from.
• A positioning consultant. Better than any skill if you can afford one, because a good consultant brings judgement about your specific market. Different price bracket entirely.
• A positioning book. Obviously Awesome and Crossing the Chasm are where this thinking comes from, and the skill is downstream of them rather than a replacement. A book cannot ask you a follow-up question when you answer "SMB owners."

The honest summary: this is the interrogation, encoded, run the same way every time. It is worth $9 if the thing you keep failing at is being specific under your own questioning — which is most people, because self-interrogation is the hardest kind.

Compatibility and install

It is plain markdown, so it runs anywhere a skill file runs. Claude Code picks it up from ~/.claude/skills/; on claude.ai you upload it to a Project; for Codex you reference it from AGENTS.md or a skills directory. Full steps ship in the QUICKSTART.

One-time $9, or $29 inside the Founder Kit alongside five other skills — demand validation, Reddit growth, X growth, a product SEO site builder, and an AI-visibility audit. No subscription; updates to the skill are included.

The field report on the product page is the skill's own working material unpacked — the niche drill and the deliverable skeleton — so you can read the exact process before deciding whether it is worth nine dollars.