| name | brand-brain-schema |
| description | Use this skill whenever the user asks Claude to set up, compile, or query a DTC brand brain. Defines the canonical raw/ → wiki/ → outputs/ structure, the six core wiki articles, the cross-linking rules, and the source-citation requirements. Activates on /compile-wiki, /scaffold-brain, and any brand-brain question. |
Brand Brain Schema
Purpose: The structural contract for Jake's DTC brand brain. Every raw file maps to a wiki article. Every wiki claim cites a raw source. Every output compounds back into the wiki on next compile.
CANONICAL STRUCTURE
<brand>-brain/
├── raw/ ← Source material (Jake writes; Claude reads)
│ ├── ads/ → fuels wiki/hooks-that-work.md + performance-patterns.md
│ ├── customers/ → fuels wiki/customer-pains.md
│ ├── competitors/ → fuels wiki/competitor-angles.md
│ ├── brand/ → fuels wiki/brand-voice.md
│ ├── performance/ → fuels wiki/performance-patterns.md
│ └── notes/ → fuels wiki/creative-brief-library.md (+ all articles)
├── wiki/ ← Synthesized intelligence (Claude maintains)
│ ├── INDEX.md
│ ├── hooks-that-work.md
│ ├── customer-pains.md
│ ├── competitor-angles.md
│ ├── brand-voice.md
│ ├── performance-patterns.md
│ └── creative-brief-library.md
└── outputs/ ← Answers and reports (Claude writes; feeds back into next compile)
├── briefs/
├── question-answers/
└── health-checks/
RAW FOLDER CONTENTS (what Jake drops in)
raw/ads/
- Meta Ads performance CSV (last 30-90 days)
- TikTok Ads performance CSV
- Top 20 performing creative screenshots
- Ad copy text files (top 10 ads' copy)
- GetHookd exports of competitor winners
raw/customers/
- Reviews export (Judge.me, Yotpo, Trustpilot)
- Customer survey responses
- Customer interview transcripts
- Support ticket themes
- Reddit threads / Amazon Q&A relevant to the niche
raw/competitors/
- Competitor ad screenshots (Meta Ad Library)
- Competitor landing page screenshots or copy dumps
- Brand positioning notes
- TrendTrack competitor briefs
raw/brand/
- Brand voice document
- Positioning statement
- Value proposition canvas
- StoryBrand BrandScript
- Avatar docs
- Necessary Beliefs doc
raw/performance/
- Weekly performance reports
- Monthly business reviews
- GA4 / Shopify analytics exports
- Email/SMS campaign performance
- LTV/CAC reports
raw/notes/
- Meeting notes (coach reviews, strategy sessions)
- Loose ideas and observations
- Hypotheses to test
- Things Jake noticed in the market
THE SIX WIKI ARTICLES (templates)
Each article follows the same structure:
# <Article Title>
**Last compiled:** <YYYY-MM-DD>
**Brand:** <name>
**Sources read:** <count of raw files this compile>
---
## Summary
[3-5 sentence executive summary]
## Key findings
[The most important insights, ranked]
## Detailed analysis
[Section-by-section breakdown with source citations]
## Cross-links
- See also: wiki/<other-article>.md for [topic]
## [VERIFY] flags
[Any claims that need human verification]
Article: hooks-that-work.md
Synthesizes from raw/ads/ and raw/performance/.
Contains:
- Top 10 hooks by performance (CTR, hook rate, conversion)
- Hook patterns (pattern interrupts, curiosity, pain-first, social proof)
- Hook + angle pairings that compound
- Failed hooks and why
- Hook types underrepresented in current testing
Article: customer-pains.md
Synthesizes from raw/customers/.
Contains:
- Top 10 pain points by frequency in reviews/support
- Internal vs external vs philosophical pain (StoryBrand schema)
- Pain points that map to existing ads
- Pain points NOT being addressed (creative whitespace)
- Customer voice samples (paraphrased, never quoted directly per Meta compliance)
Article: competitor-angles.md
Synthesizes from raw/competitors/.
Contains:
- Top competitors and their dominant angles
- Saturated angles to avoid
- Competitor weaknesses / underserved angles
- Where Jake's brand has positioning whitespace
- Competitor's evolution over time (what they've moved away from)
Article: brand-voice.md
Synthesizes from raw/brand/.
Contains:
- Brand voice attributes (tone, vocabulary, sentence structure)
- StoryBrand role definitions for this brand
- Jungian archetype mapping (if applicable)
- Drift detection — where ads/copy diverge from voice
- Voice examples (best-fit copy from
raw/ads/)
Article: performance-patterns.md
Synthesizes from raw/performance/ and raw/ads/.
Contains:
- What's correlated with winners (hook type, angle, format, length)
- What's correlated with losers
- Seasonal patterns
- Audience patterns (which demos convert best)
- Day-of-week / time-of-day patterns (if data supports)
Article: creative-brief-library.md
Synthesizes from raw/notes/ and outputs/.
Contains:
- Past creative briefs (what was planned, what shipped)
- Templates for new briefs
- Lessons from each campaign
- Open hypotheses to test
INDEX.md (auto-maintained)
# Brand Brain Index — <Brand>
**Last compiled:** <date>
**Total raw files:** <count>
**Total questions answered:** <count from outputs/>
## Articles
1. hooks-that-work.md — [last summary]
2. customer-pains.md — [last summary]
...
## Recent questions asked
- [Date] [Question slug] → outputs/<file>
- ...
## Health
- Last health check: <date>
- Contradictions outstanding: <count>
- Articles with no source citation: <count>
COMPILATION RULES
When /compile-wiki runs:
- Read every file in
raw/, top to bottom
- For each wiki article, synthesize the relevant raw files into the article structure
- Cite sources inline —
(source: raw/ads/meta-q1-export.csv, row 47)
- Cross-link aggressively — every article should link to ≥2 other articles where topics intersect
- Update INDEX.md with the new compile date and any summary changes
- Pull from
outputs/ — saved answers from prior questions feed back in
- Flag contradictions — if
raw/brand/voice.md says "premium" but raw/ads/ performance favors casual, surface that conflict in brand-voice.md
- Do NOT invent claims. If a topic isn't covered in
raw/, the wiki should say so explicitly, not guess.
QUERY RULES
When Jake asks a question against the brain:
- Read the relevant wiki articles first (always — they're the synthesis layer)
- Dip into
raw/ if the wiki doesn't have enough detail
- Cite sources in the answer — point Jake to specific files
- Save the answer to
outputs/question-answers/<slug>-<date>.md
- Flag any [VERIFY] if the answer hinges on unverified data