Claude Skills for Meta Ads: 6 Downloadable Skills for Copy, Campaigns, and Audits

Meta Ads is the channel where copy quality compounds most. A 0.4% lift in hook strength on a $50k/month ASC campaign is more dollars than the same lift on a Search campaign, and the velocity of copy testing is brutally higher than any other channel. If you ship 6 ads a week and 2 ads have to be re-written because the brand voice drifted, you’re losing real money to context-switching.

Claude skills solve this by encoding the prompt once, in a file. You install the skill, run it with a one-line invocation, and the skill follows its own internal rulebook, character limits, hook framework selection, cold-vs-warm-vs-retargeting logic, banned-claim sweeps. Paired with Claude Memory, brand voice and USPs are loaded automatically.

We’ve published six free Claude skills for Meta Ads. This post is the operator’s guide to all six.

Download the bundle →

The six skills, at a glance

All free. All standalone, they work without any API integration to Meta. All composable, they’re designed to chain in one Claude session.

1. meta-primary-text, the hook generator

Primary text is where Meta ads live or die. The first 125 characters of your primary text are what shows above the “See more” truncation on mobile, and that’s where most of the click decision happens.

What it does: Generates 8 primary text variations, each using a different hook framework:

1. PAS (Problem · Agitate · Solve)
2. AIDA (Attention · Interest · Desire · Action)
3. STORY (person + moment + arc)
4. CONTRARIAN (contradict the audience’s assumption)
5. STAT (single startling number)
6. QUESTION (audience answers “yes” in their head)
7. BEFORE/AFTER (contrast two states)
8. LIST (numbered scannable benefits)

Each variation is tagged with the framework name and the audience temperature it fits best (cold, warm, or retargeting).

Hard rules the skill enforces:

• First line ≤ 125 characters (mobile truncation)
• No fake stats (every number must be real and sourceable)
• Match audience temperature (retargeting copy doesn’t introduce the product)
• No “click the link in bio” language (this is a paid ad with a real CTA button)
• Banned-claim sweep before output

Why hook framework diversity matters: If you ship 8 variations all using PAS, you’ve tested wording, not framing. The skill forces you to test 8 frameworks so you learn what your audience actually responds to.

See the full skill →

2. meta-headline-descriptions, the supporting copy layer

The headline (under the image) and link description (under the headline) are where the verdict happens after the primary text earns the read.

What it does: Generates 10 headlines (≤40 characters), 5 link descriptions (≤30 characters), and 5 CTA button recommendations ranked by fit with the audience temperature.

Why it’s a separate skill from primary text: Headlines should extend or contrast the primary text, not repeat the hook. The skill takes the primary text as an input so it can deliberately avoid duplicating the first line.

CTA-to-temperature matching: The skill matches CTAs to audience temperature, not strongest-button-by-default. Cold audiences rarely respond to commitment-heavy CTAs (“Order Now”, “Subscribe”); they respond to softer entry CTAs (“Learn More”, “Shop Now”). Retargeting audiences are the inverse, they don’t need education, they need a push to convert.

See the full skill →

3. meta-advantage-plus-creative-pack, the ASC heavyweight

Advantage+ Shopping (ASC) lets the algorithm test combinations across up to 10 primary texts, 10 headlines, and 10 descriptions per ad. The skill produces a full pack designed for this combinatorial format.

What it does: Generates a complete creative pack:

• 5 primary text variations (5 different frameworks)
• 5 headlines (modular, each works with any primary)
• 3 link descriptions (risk reversal / logistics / proof)
• 3 CTA buttons (primary / secondary / stretch)
• 3–5 audience signals (interests / custom audience / lookalike)
• A diversity score on hook, audience coverage, and lifecycle coverage

The diversity score is the key feature. If the pack scores < 4 on any dimension, the skill rewrites the weakest variation before delivering.

Where it earns its keep:

• ASC campaign launches (cuts pack creation from ~2 hours to ~20 minutes).
• Monthly ASC refresh (you should refresh creative every 30 days; the skill makes it a small task).
• Splitting an existing ASC into two product-themed ASCs.

Hard rules:

• All 5 primaries use 5 different frameworks (no PAS twice).
• Modular headlines, no headline can reference a fact only present in one specific primary.
• All offer claims must be live on the LP.
• Maximum 1 emoji per primary text.

See the full skill →

4. meta-ad-teardown, the diagnostic skill

The hardest skill to do well manually. Looking at a losing ad and diagnosing what’s wrong requires separating signal from noise, and most teardowns conflate the ad with the audience with the offer.

What it does: Takes an underperforming ad + its metrics and outputs:

• A diagnosis matrix scoring each component (hook / body / visual / CTA / audience / LP / offer) with metric evidence and confidence (1–5)
• A primary diagnosis paragraph stating the most likely cause
• Three net-new angles to test, each addressing a different component of the diagnosis, with hypothesis, rationale, and test design (budget, success metric, decision rule)

Why metric-grounded diagnosis matters: A teardown that says “the hook is weak” is opinion. A teardown that says “0.6% CTR vs 1.2% account average and 4-second average watch time on a 15-second video → hook is weak” is evidence. The skill enforces the latter.

Common diagnosis patterns the skill recognises:

• Low CTR + low CPM → hook problem (audience saw it, didn’t click)
• Low CTR + high CPM → audience problem (Meta couldn’t find good prospects)
• High CTR + low CVR → LP mismatch or offer mismatch
• High CTR + high CVR + low ROAS → AOV / pricing problem, not creative
• Decaying CTR + high frequency → creative fatigue
• Strong day-1, dies by day-7 → audience too narrow or wrong learning signal

See the full skill →

5. meta-ads-campaign-builder, the campaign blueprint generator

For when you have a brief and need a full campaign architecture, not just an ad.

What it does: Takes a brief (business type, product, catalog status, pixel/CAPI status, geography, budget, AOV, audience assets, creative assets) and returns:

• A recommended campaign architecture (one of three patterns: ASC single-campaign, three-objective stack, two-objective stack)
• Full ad set structure with audiences, placements, optimisation events, budget, schedule
• A 3×3 creative matrix (audience temperature × angle)
• Pre-launch audience build checklist (customer list, lookalikes, retargeting layers)
• Budget split with explicit math
• A 14-day learning-phase plan, day-by-day or week-by-week
• A naming convention to apply across all generated names
• A pre-launch tracking checklist (Pixel, CAPI, EMQ, dedup, AEM, catalog)

The pattern recommendation is the most useful part. Most operators default to a 3-objective stack because that’s how they were trained. The skill recommends ASC where ASC fits and 3-stack where it doesn’t, with explicit criteria (catalog connected, ≥50 purchase events/wk, ≥$5k/mo spend).

Hard rules:

• No > 3 ad sets per campaign without explicit justification (CBO needs critical mass).
• No manual detailed-targeting narrowing on cold prospecting (2026 Meta penalises this, broad performs).
• No Cost Cap as a starter strategy.
• No > 5 creatives per ad set at launch.
• Always separate UGC and brand creatives into their own ad sets, they have different baseline CTR profiles.

See the full skill →

6. meta-ads-account-auditor, the takeover and refresh tool

Walks an 8-section audit and outputs a prioritised punch list. Order:

1. Pixel + Conversions API (always first, bad tracking poisons everything)
2. Account structure (campaign count, ad set count, ASC vs not, audience overlap)
3. Creative (count per ad set, diversity, fatigue, format coverage)
4. Copy (hook variety, headline-LP alignment, CTA-temperature match, banned claims)
5. Audiences (customer list freshness, lookalike coverage, detailed targeting on cold, excluded audiences)
6. Placements + delivery (Advantage+ Placements adoption, optimisation event fit, bid strategy)
7. Catalog (if e-com, product count, feed errors, product sets)
8. Attribution + reporting (attribution window, comparison window, holdout test status)

Every finding cites the specific campaign / ad set / metric. Every recommendation has a measurable expected outcome.

Where it’s most useful:

• Agency takeovers (60-minute first-day audit replaces a 3-day deep dive).
• Pre-Q4 prep (enter peak with no Critical issues).
• CPA-lifting projects (forces the prioritisation that ad-hoc audits skip).

A favourite hard rule: “Never recommend pausing a campaign with > 4x baseline ROAS without first checking attribution settings. It may just be receiving credit for organic.” Most LLM audits fail this check.

See the full skill →

The chained Meta workflow

How the six skills compose for a real campaign launch:

Brief (catalog connected, pixel + CAPI live, budget, AOV, audience assets)
      │
      ▼
[meta-ads-campaign-builder]            ──>  Architecture (ASC or 3-stack), audiences, naming, 14-day plan
      │
      ▼
[meta-advantage-plus-creative-pack]    ──>  Full pack: 5 primaries + 5 HL + 3 desc + 3 CTAs + signals
                  OR
[meta-primary-text] + [meta-headline-descriptions]  for non-ASC campaigns
      │
      ▼
Build in Ads Manager, launch.
      │
      ▼ (Day 7+)
For any ad underperforming baseline:
[meta-ad-teardown]                     ──>  Diagnosis + 3 net-new angles
[meta-primary-text]                    ──>  New hooks for the next test
      │
      ▼ (Day 30+ or quarterly)
[meta-ads-account-auditor]             ──>  Full punch list

The full chain runs inside one Claude conversation (or split across several). Memory is inherited so brand voice and audience profile show up in every output.

For a worked end-to-end example with real prompts and real outputs, see chaining Claude skills end-to-end.

What the skills don’t do

The skills generate copy, packs, blueprints, and audit findings. They don’t push to your Meta account. For that, pair them with:

• Meta Ads MCP, Meta’s official MCP server (launched April 2026, 29 tools, 10 of which are catalog ops). Gives Claude read-write access to your account.
• Catalog tooling, for catalog-side workflows, see the product feed management skills.

The skills work standalone. They become live-account operators when paired with MCPs.

Adoption order

If you’ve installed all six, here’s where to start:

1. meta-primary-text on a product you know well, see if the 8 framework variations beat your usual one-framework approach.
2. meta-advantage-plus-creative-pack on your next ASC refresh, replace the 2-hour write with a 20-minute review.
3. meta-ad-teardown on your worst-performing active ad, see if the diagnosis matches your gut.
4. meta-ads-account-auditor on one account, looking for Critical issues before peak.
5. meta-ads-campaign-builder on the next new-campaign brief.
6. meta-headline-descriptions as the always-on companion to your primary-text workflow.

If you also run Google Ads, see the Claude skills for Google Ads guide. If you also manage product feeds, see the product feed management skills.

Download all 16 skills →