An AI strategist that finds the creator angles, hooks, and concepts most likely to break out — before you pay to source or produce.
Works with Claude Code, OpenClaw, or any agent framework that can read skills.
Brand Truth → Breakout Patterns → Creator Archetypes → Concept Slate → Test Plan → Managed Execution
This kit automates the creator strategy layer most brands skip:
- Understand the brand — Extract buyer psychology, desire states, objections, and brand truth from the site
- Find breakout patterns — Pull post-level performance outliers, not follower count
- Map creator archetypes — Who should deliver the idea and in what filming style
- Generate test-worthy concepts — Hooks, filming beats, and angles tied to real buyer tension
- Rank the first 3 tests — What to run first, what to run alongside, what to backlog
- Call out generic plays to reject — The 5 things you would NOT do, so the strategy feels sharp
- Render a Creator Breakout Brief — A client-ready markdown + dark-terminal-aesthetic HTML artifact
The result: a short, sharp test plan you can hand into sourcing, briefing, and managed execution — without ever sounding like a generic UGC brainstorm.
👀 See the output first:
shared/examples/sample-breakout-brief.html— a complete sample Creator Breakout Brief rendered in a dark terminal aesthetic. Open it in a browser to see what a client-ready brief actually looks like.
I've spent 20 years in marketing. Scaled Fireball Whisky from one state to a billion-dollar global brand. Ran campaigns for Heineken, Hennessy, Buffalo Trace. Now I run Emerald Digital, an AI-first marketing agency.
Here's what I learned: creator budget burns fastest when you source before you strategize.
Most creator programs start the same way. Pick a short list of creators someone on the team already likes. Write a generic "authentic UGC" brief. Ship three videos. Hope. Then spend tens of thousands of dollars learning that the ad angle and the creator talent were mismatched from the start.
The problem isn't the creators. The problem isn't even the execution. The problem is that the strategy layer never existed. Nobody asked what is actually breaking out in the category. Nobody mapped the top objection to a specific hook structure. Nobody rejected the generic plays before briefing.
Creator Breakout Kit fills that gap. It's the layer that runs before sourcing — the one that turns "we need creator content" into a real test plan.
I'm open-sourcing this because every brand and every agency running paid creator programs deserves a strategist before they deserve a shopping list.
Not a brainstorm. Not a generic UGC recommendation doc. Not a follower-count spreadsheet. Not a brief that says "authentic creators, 3 hooks each."
A strategy layer before sourcing.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ Brand Truth │───▶│ Breakout │───▶│ Creator │───▶│ Concept │
│ (Firecrawl) │ │ Patterns │ │ Archetypes │ │ Slate │
│ │ │ (Virlo) │ │ │ │ │
└──────────────┘ └──────────────┘ └──────────────┘ └──────────────┘
│
┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ Managed │◀───│ Brief │◀───│ First 3 │◀─────────┘
│ Next Step │ │ (MD + HTML) │ │ Tests │
└──────────────┘ └──────────────┘ └──────────────┘
Every stage reads from the previous stage's artifacts. No stage hallucinates inputs. No stage ships without the honest evidence label.
- Outlier logic, not follower vanity — Patterns come from post-level breakout ratios vs a creator's own baseline, not from "top 10 beauty creators on TikTok"
- Buyer psychology drives the concept — Every concept maps to a specific objection or desire state, not a generic "authentic UGC" angle
- Named-framework labels — Every concept card carries Breakout Driver · Buyer Tension · Creator Delivery Style · Test Priority so the strategy feels ownable, not vibes
- Score strip per concept — Strategic fit · Breakout evidence · Production ease · Confidence, computed from research mode and source-post presence (Strong directional / Directional / Hypothesis only)
- Anti-positioning block — Every brief explicitly lists the 5 generic plays being rejected, which sharpens the strategy and blocks scope creep
- Evidence discipline — Virlo metadata is labeled metadata-only until video intelligence runs; audience sentiment is labeled directional, never "market research"
- Creator names stay comps — Never a roster. Never "already lined up." Never "ready to post." Until sourcing has actually happened
- Two render formats from one data source — Markdown for editing and agent consumption, HTML for visual presentation. Both renderers share score computation via
importlibso they can't drift
Every synthesized run produces a brief with this exact structure — in both creator-breakout-brief.md and creator-breakout-brief.html:
| # | Section | What it delivers |
|---|---|---|
| 1 | Executive Summary | One-page punch: three standout concept directions named explicitly, the move, what we would ship first, evidence base |
| 2 | The Breakout Thesis | Category insight — what is actually driving breakout in this space right now |
| 3 | Pattern Board | 3–5 named breakout patterns with Breakout Driver, Reusable Mechanic, How-to-use-it, source references |
| 4 | Creator Archetypes | Role-based cards (not a roster): role in the feed, energy, why they fit the pattern, brief type |
| 5 | Concept Slate | Concept cards with the full framework labels and animated score strip |
| 6 | First 3 Tests We Would Run | Numbered list with one-sentence why, test shape, decision window, success signals |
| 7 | What We Would NOT Do | Anti-positioning block — the 5 generic plays explicitly rejected |
| 8 | Managed Next Step | Crisp handoff into sourcing, briefing, and managed execution |
Plus an Appendix: Brand and Audience Context with product, audience, objections, hero products, brand truths, and claim/brand-safety guardrails.
| Skill | What it does |
|---|---|
creator-breakout-kit |
Orchestrator — runs the full pipeline end-to-end and produces the Creator Breakout Brief |
brand-snapshot |
Extract buyer psychology, desire states, objections, and brand truth from site source material |
breakout-pattern-research |
Identify repeatable breakout patterns from post-level outlier behavior, not follower count |
creator-archetype-mapper |
Translate winning patterns into creator roles (educator, skeptic, routine fixer) — comps only, never a roster |
creator-concept-generator |
Build the concept slate with named-framework labels, hooks, filming beats, and guardrails |
creator-kit-formatter |
Render the structured kit into the Creator Breakout Brief + internal notes |
sales-handoff |
Optional bridge — turn the brief into honest sendable follow-up copy |
Each skill is a plain markdown file. Any agent framework that reads skills can run this kit.
Generic creator brief:
"Need 5 lifestyle creators to make authentic videos about our product."
Creator Breakout Kit direction:
"Anchor with an outcome-proof test (evidence-led explainer, honest caveat). Run alongside an objection-reset concept from a practical skeptic resolving the top purchase concern on-camera. Backlog a real-use demonstration from a creator who shows the product in existing behavior without extra production. Skip polished lifestyle montages — the category rewards credible discovery."
That is the difference between buying production and buying a test plan.
| Typical workflow | Creator Breakout Kit |
|---|---|
| Find creators first | Find breakout concepts first |
| Pick based on audience/follower fit | Pick based on pattern fit + buyer fit |
| Brief generic UGC | Brief specific hooks, structures, and angles |
| Hope the content works | Start from what is already breaking out |
| Burn budget learning late | Learn strategically before sourcing |
- Agencies selling creator strategy or managed UGC
- Brands that want stronger creator briefs before production
- Paid social teams who need better creator angles, not just more creators
- Operators building a strategy-first creator sourcing pipeline
git clone https://github.com/themattberman/creator-breakout-kit.git
cd creator-breakout-kit
cp .env.example .env # fill in only the connector keys you have
bash scripts/preflight.sh # verify what modes are availableThe kit runs in quick-pass mode with zero connectors. Add Firecrawl + Virlo for outlier-pass (the recommended default). Add ScrapeCreators + OpenRouter for the optional video-intelligence stage.
RUN_DIR="runs/$(date +%F)-example-brand"
RESEARCH_DIR="${RUN_DIR}-outlier-pass"
python3 scripts/create_base_breakout_run.py \
--brand-name "Example Brand" \
--brand-url "https://example.com" \
--product-offer "A hero product or offer customers can understand quickly" \
--audience "The specific buyer this creator test should reach" \
--primary-desire "The outcome the buyer wants most" \
--primary-objections "too expensive,not sure it works,already have an alternative" \
--platform-focus "TikTok + Reels" \
--run-dir "$RUN_DIR"
bash scripts/run_creator_breakout_research.sh \
--brand-name "Example Brand" \
--brand-url "https://example.com" \
--niche "example category" \
--keywords "example keyword,buyer objection,category trend" \
--out-dir "$RESEARCH_DIR"
python3 scripts/enrich_breakout_run.py \
--run-dir "$RUN_DIR" \
--research-dir "$RESEARCH_DIR" \
--overwrite
python3 scripts/synthesize_full_kit.py --run-dir "$RUN_DIR"After synthesize_full_kit.py runs, $RUN_DIR contains both creator-breakout-brief.md and creator-breakout-brief.html. Open the HTML in a browser to see the show-off version.
# Audience sentiment from Virlo Tracking (comment-derived directional signal)
python3 skills/breakout-pattern-research/scripts/virlo_tracking_intelligence.py \
--research-dir "$RESEARCH_DIR" --limit 3 --track videos --scrape-cadence weekly
python3 scripts/enrich_breakout_run.py --run-dir "$RUN_DIR" --research-dir "$RESEARCH_DIR" --overwrite
# Video intelligence (ScrapeCreators + Gemini via OpenRouter) — upgrades evidence from metadata-only
python3 skills/breakout-pattern-research/scripts/video_intelligence.py --run-dir "$RUN_DIR" --limit 3 --max-video-mb 20
python3 scripts/synthesize_full_kit.py --run-dir "$RUN_DIR"Virlo Tracking is opt-in because tracking cycles can incur Virlo costs. The first run may only queue tracking; rerun the adapter later to pull ready reports.
| Mode | Connectors | When to use |
|---|---|---|
quick-pass |
None | Fast internal draft, no connectors needed |
brand-pass |
Firecrawl only | Brand-informed draft without outlier evidence |
pending-data |
Firecrawl + Virlo (queued) | Connectors started but evidence not yet populated |
outlier-pass |
Firecrawl + Virlo | Recommended default when the brief is going to a stakeholder |
full-kit-synthesis |
Post-research synthesis | Materializes the final brief and stage JSON |
deep-pass |
Outlier-pass + audience signals + video intelligence | Paid strategy asset or high-value engagement |
Audience signals (Virlo Tracking) and video intelligence (ScrapeCreators + Gemini) are optional stages that layer on top of outlier-pass. They do not define their own mode label — they add evidence to an existing run.
Missing connectors degrade the mode label honestly. They don't silently degrade into fake certainty.
Hard rules baked into every stage:
- Never imply creators are booked, sourced, or committed unless that is explicitly true. Creator names are reference comps only.
- Virlo metadata is metadata-only unless
video-intelligence.jsonexists for that post withvideo_asset_sent_to_model: true. Do not infer visual beats from captions. - Virlo Tracking output is directional signal, not statistically representative market research. Label it that way in every output.
- An outlier-pass run is not client-ready until
synthesize_full_kit.pyhas materialized the final brief. Keep that gate intact. - Missing connectors downgrade the mode label (quick-pass / brand-pass / pending-data) rather than filling gaps with vibes.
Every standard run produces:
creator-breakout-brief.md— the Creator Breakout Briefcreator-breakout-brief.html— dark-terminal-aesthetic HTML version (synthesis runs only)internal_notes.mdbreakout-kit.json- optional
sales_handoff.md
Connector-backed research can also produce:
brand-discovery.json/brand-discovery.mdfirecrawl-pages.jsonvirlo-outlier-research.json/virlo-outlier-research.md- optional
virlo-audience-signals.json/virlo-audience-signals.md
Full-kit synthesis also materializes:
brand-snapshot.jsonbreakout-pattern-research.jsoncreator-archetype-map.jsonconcept-slate.jsonskill-run-manifest.json
Optional video intelligence can also produce:
video-intelligence.json/video-intelligence.md
This isn't about making one brief faster. It's about making the strategy layer cheap enough to run before every creator program, not after it fails.
Before: "We need creator content" → generic UGC recommendations deck → tens of thousands of dollars spent on sourcing and production built on vibes → underperformance → post-mortem
After: "We need creator content" → Creator Breakout Brief (markdown + HTML) in 10 minutes → three test-worthy concepts with honest confidence labels → managed sourcing and briefing runs against a real test plan → decisions come from signal, not hope
Run this alongside the rest of the kit set for a full campaign loop:
- Meta Ads Kit — Daily monitoring, fatigue detection, winner scaling, ad copy generation matched to creatives
- Google Ads Copilot — Intent-first account management through Read → Draft → Apply layers
- Landing Page Factory — Turn any URL into a deployable landing page matched to the ad angle
Creator Breakout Kit is the strategy layer that feeds all three: the brief decides what to test, the other kits run and measure it.
creator-breakout-kit/
├── README.md # You're here
├── CLAUDE.md # Architecture notes for Claude Code instances
├── docs/
│ ├── integration-notes.md
│ └── setup.md
├── skills/ # ← THE CORE (works with any agent)
│ ├── creator-breakout-kit/ # Orchestrator
│ ├── brand-snapshot/ # Buyer psychology + brand truth
│ │ ├── SKILL.md
│ │ ├── scripts/ # firecrawl_brand_discovery.py
│ │ └── references/ # firecrawl-brand-discovery.md spec
│ ├── breakout-pattern-research/ # Post-level outlier research
│ │ ├── SKILL.md
│ │ ├── scripts/ # virlo_outlier_research.py, virlo_tracking_intelligence.py, video_intelligence.py
│ │ └── references/ # virlo-outlier-research.md spec
│ ├── creator-archetype-mapper/ # Role-based archetypes
│ ├── creator-concept-generator/ # Concept slate with named frameworks
│ ├── creator-kit-formatter/ # Client + internal deliverables
│ │ ├── SKILL.md
│ │ └── references/ # client-brief-template.md, internal-notes-template.md
│ └── sales-handoff/ # Optional handoff copy
├── shared/
│ ├── schemas/ # breakout-kit.schema.json
│ ├── templates/ # style-profile.md (cross-skill)
│ └── examples/
│ ├── example-invocation.md
│ └── sample-breakout-brief.html # ← the dark-terminal show-off asset
├── scripts/ # Cross-skill pipeline orchestration
│ ├── preflight.sh
│ ├── run_creator_breakout_research.sh
│ ├── create_base_breakout_run.py
│ ├── enrich_breakout_run.py
│ ├── synthesize_full_kit.py
│ └── render_breakout_brief_html.py # ← parametric HTML renderer
├── tests/
├── evals/
├── .env.example
└── .gitignore
Run outputs are written to runs/ locally and ignored by git.
| Tool | Monthly Cost |
|---|---|
| Creator Breakout Kit skills | Free forever (MIT) |
| Claude Code | Included with Claude Pro/Team subscription |
| Firecrawl | Free tier available |
| Virlo | Paid (required for outlier-pass) |
| ScrapeCreators + OpenRouter | Pay-as-you-go (optional, video intelligence only) |
| Kit itself | $0/mo |
The kit runs in quick-pass mode with zero connectors if you're not ready to pay for Virlo yet. You still get a brief — just with Hypothesis-only confidence labels instead of Strong directional.
Before this ships publicly, make sure:
.env.examplelists required and optional connector keys without private defaultsscripts/preflight.shexplains which modes are available- Missing connectors produce honest degraded-mode labels
- Generated run outputs stay local and are not checked into git
- A new operator can run the README path without reading prior run history
A strong Creator Breakout Brief:
- Feels brand-specific, not agency-generic
- Uses outlier logic instead of follower vanity
- Maps concepts to real buyer objections and desire states
- Stays honest about evidence limits and creator availability
- Makes the next paid step feel obvious without sounding pushy
MIT License. See LICENSE. Use it, fork it, build on it.
Built by Matt Berman.
- 🐦 Twitter/X: @themattberman
- 📰 Newsletter: Big Players
- 🏢 Agency: Emerald Digital
Stop guessing which creator concepts to test. Let breakout patterns, buyer psychology, and brand truth do the picking — before you pay to source or produce.
Star the repo if this helps. It tells me to keep building.