What's in the toolkit
A skill is the visible part. Behind each one sits a reference layer the skill loads on demand: standards it enforces, libraries it classifies against, templates it copies. You do not call any of this directly, but knowing it exists explains a lot of skill behavior.
Nothing here changes when you update the toolkit’s brand data, and nothing here is yours to edit. Your files live somewhere else entirely.
skills/ — the 44 skills plus 2 bundled tools
Section titled “skills/ — the 44 skills plus 2 bundled tools”Each is a folder with a SKILL.md (always loaded: identity, hard rules, workflow) and a references/ folder loaded only when needed. That split is why a skill can carry a great deal of methodology without consuming your context on every run.
The full catalog is on the skills overview.
rules/ — the standards every skill obeys
Section titled “rules/ — the standards every skill obeys”These fire automatically. They are the reason output is consistent across skills rather than depending on which one you happened to call.
| Rule file | What it governs |
|---|---|
preflight-gate.md | Blocks output until minimum brand data exists. This is what stops a skill producing confident generic work about a brand it knows nothing about |
verdict-grammar.md | The four verdict values and why NO-DATA blocks |
pull-ladder.md | How a number gets sourced, and account binding |
conversion-shapes.md | Your declared money path |
state-conventions.md | Where durable state is written |
setup-message-contract.md | The five parts of a “setup needed” message |
design-standards.md | Visual standards for every page type: banned patterns, visual density, animation |
advertorial-standards.md | Advertorial and landing-page copy and design rules |
ad-creative-standards.md | Ad creative: aspect ratios, readability, text rendering, video rules |
voice-authentication.md | Voice matching and AI-tell detection |
quiz-funnels.md | Quiz funnel requirements |
html-deployment.md | Paths, QA checklist, the automated detectors |
context-management.md | How long workflows hand off across sessions |
skill-standards.md | Architecture standards for skills themselves |
knowledge-base/ — 33 reference files loaded on demand
Section titled “knowledge-base/ — 33 reference files loaded on demand”The deep material skills pull from when the task calls for it.
- Frameworks: CRO psychology, Blair Warren persuasion, the creative strategist curriculum, the DR improvement system
- Design: the design reference library, animation components with copy-paste blocks, a live animation demo, premium fonts with CDN links, landing page design
- Page types: per-type specs, so a listicle is built to listicle rules rather than generic page rules
- Craft cards: the human-judgment layer on taste and restraint
- Voice profiles: writing-voice references
- Platform and compliance: ad platform specs, compliance guardrails, authentic AI people prompting
- Imagery: the Nano Banana model guide, ad style examples, ad swipe references
- SEO: the programmatic SEO playbook, AI and LLM search optimization
ad-brain/ — the creative intelligence library
Section titled “ad-brain/ — the creative intelligence library”What every ad skill classifies against. This is why ad output references named approaches rather than improvising a category each time.
| File | What it holds |
|---|---|
approaches.md | 33 strategic ad approaches: social proof, pain point, UGC, whiteboard, comparison, contrarian and more |
visual-styles.md | 50 visual execution styles across 6 categories |
cognitive-biases.md | 34 biases plus 5 power stacks |
andromeda-rules.md | Meta’s diversity and scaling behavior |
video-formats.md | 10 video ad formats with duration and placement guidance |
rules.md | Ad creative rules distilled from production |
naming-convention.md | Approach, format, ratio, timestamp |
winners-template.md · anti-patterns-template.md | Where your best and worst performers accumulate |
Those last two are templates on purpose. They start empty and fill up with your account’s history, which is what makes the library get better the longer you use it.
templates/ — starting points for a new brand
Section titled “templates/ — starting points for a new brand”The 11 brand data file templates, plus a workspace CLAUDE-template.md and a setup guide. Copied into your brand root during onboarding, then populated. You will rarely touch these directly, since brand-context and onboard-client handle the copying.
evaluation/ — scoring rubrics
Section titled “evaluation/ — scoring rubrics”ad-creative-rubric.md: 9 dimensions for scoring an ad image, including thumb-stop power, copy hierarchy, emotional trigger, CTA strength and originality.bias-choice-matrix.md: maps awareness level × traffic temperature × category to the right psychological lever, so bias selection is a lookup rather than a hunch.golden-tasks/: fixture brands used to test skills cold before release.
agents/ and examples/
Section titled “agents/ and examples/”agents/ holds reusable behavior templates skills delegate to. examples/ holds completed outputs using a fictional brand, so you can see the quality bar before you run anything.
What this means for you day to day
Section titled “What this means for you day to day”- You do not load any of this manually. Skills pull what they need.
- It explains “why did it do that.” A skill refusing a layout or an aspect ratio is usually enforcing a rule file, and the rule file says why.
- Two ad-brain files are yours to grow. Winners and anti-patterns accumulate from your own results.
- Updating the toolkit updates all of it at once, and never touches your brand data.