skill
grill-my-ads
A relentless media buyer in your terminal: Claude skill that grills you about your product, reads your repo, and builds Meta Ads (FB/IG/WhatsApp) via the official Meta Ads MCP — drafts first, publish
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/Andretuta/grill-my-ads itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 22, 2026Why
Most ad money is lost before the first impression: wrong objective, no margin math, a homepage as destination, five copies of the same creative. grill-my-ads interviews you like a senior media buyer before spending a cent, does the research itself, and only then builds the campaign.
What it does
| Command | What happens |
|---|---|
/grill-my-ads | Grills you in rounds (numbered questions, each with a recommended answer) → researches the most viable objective/destination → writes 3–5 genuinely different concepts → uses your photos/videos or writes a UGC video script → builds campaign, ad set, creatives and ads as drafts → shows previews → publishes only on explicit "ok" |
/grill-my-ads --repo | Same flow, but first reads the current project (landing page, pricing, routes, real user flow) to prefill the interview and render a product video from the code (via brag). Only when you ask — the default never touches your files |
/grill-my-ads audit | Read-only account audit: ~40 checks (pixel/CAPI, creative diversity & fatigue, structure, audience, compliance) → 0–100 score, A–F grade, quick wins |
/grill-my-ads roi | Joins Meta spend with your business data (orders, qualified leads, retained members) → real CPA, ROAS vs breakeven ROAS, profit, and a SCALE / KEEP / CUT verdict per ad |
Highlights
- Facts it looks up, decisions you make. Account, Page, IG, pixel, history → fetched via MCP. Budget, offer, approval → asked.
- Works for any product. The interview alone is enough — coffee, courses, local services, SaaS.
- Repo → ad, opt-in. With
--repoit reads your landing page, pricing, routes and real user flow; never.env, keys or secrets. Without it, it never reads your files. - Video creatives. Uses your media, writes a UGC script, or (with
--repo) renders a 9:16 + 4:5 product video from the codebase. - Safety first. Draft/PAUSED by default, pause-never-delete, never infers budget, never invents interest IDs, ≤20% budget steps, policy linter on every line of copy, AI-disclosure always asked.
- Up to date (2026). Advantage+ unification, Andromeda creative diversity, Jan-2026 attribution window removal, Brazil's 12.15% ad-tax pass-through, LGPD/GDPR notes.
Requirements
- Claude Code or Claude.ai (Customize → Skills)
- The official Meta Ads MCP connected:
https://mcp.facebook.com/ads - Python 3.9+ (for the helper scripts, stdlib only)
- Optional, for video: Node 18+, ffmpeg and the brag skill
Install
Claude Code
git clone https://github.com/Andretuta/grill-my-ads ~/.claude/skills/grill-my-ads
Claude.ai — download the repo as ZIP and upload it in Customize → Skills.
Then just say /grill-my-ads (or "grill me about my ad"). Inside a code project, use /grill-my-ads --repo to build the ad from the code.
Helper scripts
All stdlib Python, each with --selftest.
| Script | Example |
|---|---|
scripts/budget.py | python scripts/budget.py min --cpa 40 → min daily budget to exit learning, in cents too |
scripts/policy_lint.py | python scripts/policy_lint.py --file copy.txt → flags personal attributes, guarantees, before/after… (EN + PT) |
scripts/audit_score.py | python scripts/audit_score.py checks.json → weighted score + grade + quick wins |
scripts/roi.py | python scripts/roi.py report --meta meta.csv --biz biz.csv --margin 0.45 --tax 0.1215 |
scripts/utm.py | python scripts/utm.py example.com/offer → standard Meta UTM template |
Layout
SKILL.md router + create flow + non-negotiable rules
references/ MCP catalog, interview tree, viability tree, creative, policy,
tracking, structure & scaling, audit, ROI, regional notes, sources
scripts/ budget · policy_lint · audit_score · roi · utm
assets/ product-context, brief and video-script templates
examples/ full sample sessions + sample inputs/outputs
Examples
See examples/: a full grill session (with budget push-back and a viability table), a repo-mode session with video generation, plus ROI, audit and policy-lint inputs with expected outputs.
Disclaimer
Not affiliated with Meta. Benchmarks are ranges tagged as official [OF] or practitioner [PR]; they are not guarantees. You are responsible for your ad spend, policy compliance and privacy obligations. The policy linter is a heuristic, not an approval.
Credits
Built on ideas from Digitizers/meta-ads-mcp (MIT-0), claude-ads (MIT), latent-spaces/brag (MIT), Matt Pocock's grilling technique and others. See LICENSES.md.