Nine Claude skills for B2B product marketing teams. All nine use one shared brand kit, so you define your positioning once, and everything is written as your company.
The idea
Skills own method. The brand kit owns content. A skill knows how to run a white space analysis, how to score a draft for AI tells, or how to classify a search term by buyer intent. It does not know your company name, your named traps, your buyer, or the four things your product deliberately does not do. Those live in five markdown files, and every skill reads them before it writes a word.
That separation is the only reason a marketing skill works for anyone but the person who wrote it. Without it you get one of two failures. A generic skill produces competent copy about no company in particular. A skill with the positioning baked into the prompt works beautifully once, and then updating a tagline means editing nine files and missing three.
One thing this repo does not hand you: a differentiation framework. It cannot. Pillars derived from one company’s tradeoffs turn into adjectives the moment you apply them to another, and borrowed frameworks produce copy that reads fine and argues for nothing. What you get is the method for building your own, plus the skills that run on it once it exists.
The nine skills
brand-kit-setup
Interviews you and writes the five brand-kit files. Run this first.
Writes the kit rather than reading it.
house-style
Rewrites, audits, or scores any draft against your voice. Catches off-style patterns, banned words, and claims your capabilities file does not support. Runs as the final pass after any other skill.
Reads voice.md and capabilities.md, plus the writing templates and messaging patterns.
content-writer
Long-form assets from the kit: articles, landing pages, feature pages, emails, case studies, whitepapers, executive briefs, pillar and cluster pages.
Reads positioning.md, voice.md, brand.md, capabilities.md.
transcript-to-article
Turns a podcast, webinar, interview, or messy meeting notes into a clean transcript, a finished article, a distribution package, social assets, and a scan report.
Reads four kit files, plus the AEO style guide.
social-generator
Platform-native captions plus the text overlay, graphic headline, and paired image prompts that ship with them.
Reads brand.md, voice.md, positioning.md, capabilities.md.
company-recon
A seven-section read on any B2B company from a single URL, assembled in minutes from its public web presence: what it sells, who it says it sells to, how it is positioned, and what marketing it is visibly running. The prospect, partner, competitor, or noise call is a starting judgment for you to confirm, drawn from a website.
Reads all five.
competitive-white-space
Maps what every competitor is already saying, isolates what none of them are saying, and turns the gap into a positioning direction and stage-by-stage campaign hooks.
Reads icp.md and positioning.md, and all five for the later sections.
google-ads-audit
Classifies every search term by buyer intent as KEEP, WATCH, or CUT, quantifies wasted spend, and flags high-intent terms that are budget-constrained.
Reads icp.md: the fit matrix, search terms, and channel economics.
linkedin-ads-audit
A 35-point audit that computes an ICP Fit Score: the share of paid social spend that actually reached the right seniority, function, and company size.
Reads icp.md: the fit matrix, buying committee, and channel economics.
Every skill works on its own. Cross-references between them are optional.
Four of them chain
Each one is measurably sharper with the previous one’s context in the conversation. None of it is required, but a full pass takes an afternoon rather than a quarter.
01. company-recon
Establishes what a company sells, who it sells to, and how it is positioned, from public sources.
02. competitive-white-space
Takes that read across a competitor set and isolates the territory nobody is standing on.
03. google-ads-audit
Judges search terms against a market it now understands. High intent is a specific judgment, not an abstract one.
04. linkedin-ads-audit
Runs last because it inherits all of it, plus the demand gap the search audit found.
What you need before anything works
Five skills run in a plain Claude Code session with nothing attached: house-style, content-writer, transcript-to-article, social-generator, and brand-kit-setup. The other four have real dependencies, and each fails a different way when the dependency is missing.
Web fetch and web search
company-recon and competitive-white-space read public sources and nothing else. Without both tools in the session they have nothing to read and produce nothing. You can paste page text in yourself, but then you are doing the research and the skill is doing the analysis.
A docx skill in the session
company-recon, competitive-white-space, and both ads audits deliver a formatted Word document when a docx skill is present to build the file. Without one they fall back to markdown with every table intact, and say so. The analysis is identical either way. The deliverable is not, which matters if you were planning to send it to a CFO.
Campaign data for the ads audits
Neither audit invents numbers. Give it a marketing data connector that reads the ad account, or pull the exports yourself. Google needs the campaign, search terms, and keywords reports over the same trailing period. LinkedIn needs campaign performance plus the demographic breakdowns by seniority, company size, and job function.
Check the LinkedIn demographic export for a spend column before you rely on it. Campaign Manager’s default export often carries impressions and clicks without cost, and the ICP Fit Score is spend-weighted. The audit falls back to impression-weighted figures and labels them, which is a weaker answer honestly described.
No skill here writes to an ad platform, a CMS, or a CRM. Everything is read and analyze.
The brand kit
Five markdown files. Skills cite them by heading name, so you can reorder freely. Do not rename or delete a heading. A skill looking for What We Do Not Do and finding nothing will make claims you did not authorize.
- brand.md Identity, palette, and CTAs.
- voice.md House style and banned constructions.
- positioning.md The frame, the pillars, the named traps, and every number you are willing to publish.
- icp.md Who you sell to, in three tiers precise enough to score a campaign against.
- capabilities.md What you do, what you do not do, and which page to link when.
voice.md ships filled in rather than blank. Those defaults and the 23 weighted patterns in its scoring rubric come from one working editor’s judgment about what damages credibility with a technical B2B reader. They are one writer’s voice, and they are meant to be replaced with yours.
That rubric scores conformance to a house style. It is not an AI detector. It catches most machine output as a side effect, because machine output is unstyled by default, and it does not identify machine authorship.
A blind calibration study ran three times, ending with 23 pieces and a third scorer. It partly disconfirmed the rubric. Three independent scorers reading the same drafts disagreed by a mean of 36 points on the eight hardest pieces, while the band assignment and the rank ordering held. That makes it a triage instrument rather than a measuring one. It answers which band a draft sits in and whether this draft beats that one. It does not answer what a draft scores, so treat a published figure like “scored 72” as false precision.
Install it
/plugin marketplace add thescottking/b2b-pmm-skills
/plugin install b2b-pmm-skillsThen run:
brand-kit-setupThe skills are plain markdown. There is nothing to build.
How long the kit takes depends on one thing: whether your positioning is already decided. If it is, budget 45 to 90 minutes and treat the session as transcription. You know your frame, you can name the failures your buyers live with, and someone can pull your cost per acquisition and your platform targeting values while you talk.
If it is not decided, budget several sittings across a week or two, and expect the positioning work to happen outside the interview. Most first-pass frameworks fail the test of whether a competitor could adopt them unchanged, including ones that took a quarter and an agency to produce. Two passes is normal, and no interview compresses that into an hour.
A thin kit produces no errors, which is the problem. A missing heading stops a skill and tells you so. Three exclusions under What We Do Not Do instead of ten means a draft will describe your whole category as though you shipped it. When brand-kit-setup reports which sections came out thin, that report is the deliverable.
Or copy the worked example in examples/ampfield/ and watch the toolkit run before you fill in anything.
What is not here
- No SEO tooling. No keyword research, rank tracking, or Search Console integration.
- No CMS page builder. The skills produce copy and formatted documents, not published pages.
- No workflow or diagram generation. Nothing here animates, visualizes, or renders a process.
- No CRM or marketing automation integration. No list building, lead scoring, or sequence deployment.
- No image generation. The social skill writes image prompts. You run them wherever you run them.
- No analytics or attribution reporting. The two ads audits judge campaign exports against your ICP. They are not a reporting layer.
- Two channels, not eight. This repo audits Google Ads and LinkedIn Ads. There is no email program here, no event motion, no partner marketing, and no community work.
- No planning layer. Nothing builds a campaign plan, a content calendar, a budget allocation, or a quarterly roadmap. Deciding what to make, in what order, against what number, is still yours.
- No sales enablement output. No battlecards, objection handling documents, one-pagers, discovery question sets, or call scripts.
Some of these may show up later. None of them are in this release.