AI Marketing Panel skill
A skill that builds a panel of 5 to 15 composite customers and tests a landing page, price or email before launch.
- 58 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skill reads your customer material and runs local validation scripts; it sends nothing out itself.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/ai-marketing-panelDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill reads your customer material and runs local validation scripts; it sends nothing out itself.
This is third-party code. Review the repository files before installing.
What it does
You describe a real decision: should a landing page go live, does the price make sense, will the email get opened. The kit builds five to fifteen composite customer personas from evidence you already have: support tickets, sales calls, reviews and the reasons people turned you down. Each persona gets a long sourced dossier. Then each persona reads your draft alone, in its own context, without seeing what the others said. You get separate reactions, a synthesis that keeps disagreements intact and a scored verdict you can argue with. The skill for Claude or Codex is paired with TypeScript validators that decide whether a panel and a run meet the standard. The model reads and writes, the scripts check.
Who it is for. Business owners and small teams who write their own customer-facing copy and have nobody to show it to before it ships.
Good fit when
- A landing page is ready and you want objections before launching ads.
- Before a price rise you need to see how different customer groups react.
- A mailing needs a check of subject line and first screen.
- Reviews and call notes have piled up and can seed customer personas.
Not a fit when
- You need real customers and interviews: the panel gives direction, not proof.
- You have no customer material to build personas from.
- Node.js is older than 22.18.0 and cannot be upgraded: the validators need it.
Example request
Build a panel of eight customers from these reviews and test the new landing page on it.Limitations
This is directional, not a real test: the authors repeat that composite personas stand in for customers only as a guide. Quality depends on the evidence behind the dossiers, so without real tickets and reviews personas come out generic. The validators need Node.js 22.18.0 or newer while `room.sh` needs only `bash`. No release asset ships an archive; get the kit as a GitHub ZIP. Do not run the kit in place: make an owned copy for each use or panels and runs will mix. Read `START-HERE.md` and `INSTALL.md` first.
How to disable. Remove the skill from the Claude app or the Codex directory. Panel and run folders live in your copy of the kit and are deleted by hand.
FAQ
How do I install the skill?
Clone the repository, check Node.js with `check-node.sh`, then drag the kit ZIP into a conversation in the Claude app or point Codex at the folder. The full procedure is in `INSTALL.md`.
How do I run the validators by hand?
Create a panel with `node skills/synthetic-customer-panel/scripts/init-panel.ts --owned-root ./panels ./panels/my-panel --name "My panel"`, then check it with `validate-panel.ts`.
What do I tell the agent?
Ask it to build a customer panel or to run the "wash this artifact" check.
How is a panel different from a focus group?
It is a set of composite personas built from your data that answers fast and cheap but stays directional: it does not replace real customers.
Do personas see each other's answers?
No. Each persona reads the draft alone in its own context.
Discussion
No questions or reviews yet.
Related
SkillsA skill collection for Claude Code covering marketing and sales: content, SEO, outbound, CRO, podcasts, video and cost review
AI Marketing Suite
AI Marketing Suite for Claude Code
A set of marketing skills for Claude Code: site audit, copywriting, emails, content calendars, competitor analysis and client reports
A skill of the author's marketing principles and playbooks that the agent reads before writing copy, emails and funnels
One skill with 14 modules for a marketer: a 0–100 site audit, copy, hooks, ads, GEO, email and launches
Skills for positioning, content, LinkedIn, cold outreach and landing page audits with quick and deep modes
24 marketing skills: Google, Meta and TikTok ads, SEO, social media, cold email and analytics through the Hyper MCP.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| AI Marketing Panel (this card) | no | 58 | |
| Hiring Panel Skill | no | 294 | seven roles try to reject a resume; here a panel of customer personas judges a landing page, price or email |
| Marketing Skills for AI agents | no | 59,933 | a set of general marketing skills; this one tests a single artifact on a customer panel |
README in short
The repository holds the `synthetic-customer-panel` skill, deterministic TypeScript validators and operating docs: a start guide, an install contract and an operations manual.
README badge
Original SKILL.md
--- name: synthetic-customer-panel description: Build, run, calibrate, and maintain evidence-grounded synthetic customer panels with 5-15 explicit seats, rich persona dossiers, isolated reactions, deliberate wrong-fit dissent, separate synthesis and scoring, and rewrite/retest loops. Use when the user says build a customer panel, synthetic focus group, audience panel, buyer panel, wash this artifact, test this copy with customers, run the panel, rewrite and retest, calibrate the panel, or update panelists from new customer evidence. --- # Synthetic Customer Panel Create a customer-research instrument, not a cast of agreeable fictional characters. Build every panel in an owned workspace copied from the neutral kit. Keep the reusable base and reference PoCs unchanged during real use. ## Route the request - Build or rebuild a panel: read `workflows/build-panel.md`. - Review one artifact or sequence: read `workflows/wash-artifact.md`. - Improve a weak artifact and test it again: read `workflows/rewrite-and-retest.md`. - Add real outcomes or tune the instrument: read `workflows/calibrate-panel.md`. - Refresh seats from new evidence: read `workflows/maintain-panel.md`. - Rerun a panel on something new, swap seats out, or run several named panels at once: read `workflows/room-lifecycle.md`. Triggers include "run it again", "same panel, different concept", "clean the room", "bounce these seats", "make six new ones from new criteria", and "run five panels in parallel". Read only the references named by the selected workflow. Do not load every reference by default. ## Non-negotiable invariants 1. Work in an owned panel workspace or copied kit. Never run or modify a PoC or canonical base in place. 2. Use an explicit `panel.json` manifest. Only its active seats run. A folder scan never decides the roster. 3. Keep 5-15 active seats. Right-size from actual customer heterogeneity; do not choose 13 merely because the source video used 13. 4. Give every seat its own evidence- and assumption-labeled dossier. Target 1,200-1,800 words; validation fails below 900 words, for thin sections, repeated filler, or near-identical dossiers. 5. Model composites, not invented claims about named real people. Separate