Meta Ad Copy Sets . Agency OS

What it does

Produces the four text fields a Meta ad needs: primary text, headline, description and the CTA button choice, matched to one funnel stage. TOF copy stops the scroll, MOF copy generates leads and builds trust, BOF copy closes. It selects a hook formula and a body structure from frameworks used by Hormozi, Kern, Billy Gene and Laurel Portie, respects the visible character limits, and formats the output for where it is going: Meta Ads Manager for your agency's ads, GHL AI Ads for a client's.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

/write-ads

> write Meta ad copy

> Facebook ad headlines

> ad hooks

> retargeting copy

> TOF ad copy

When to reach for it

Writing Facebook or Instagram ad copy
Ad hooks and scroll-stopping headlines
Retargeting copy for people who already visited
Lead gen or conversion ad copy at a named funnel stage
Ads for your own agency: case study, free audit, authority
Ad copy for a client business you run ads for

When NOT to use it

If you actually want	Use this instead
The image or video behind the ad, meaning visual concepts and generation prompts	`creative-director`
Non-ad persuasion assets: emails, sales pages, sales letters, organic posts	`legendary-copywriters`
Organic social captions written from a video transcript	`social-caption-writer`
A VSL script the ad drives traffic to	`vsl-architect`
Campaign structure, audiences, budgets and Lead Form plumbing this copy loads into	`meta-lead-system`
The landing page the ad clicks through to	`funnel-architect`

Before you start

What you need	Why
A decided funnel stage: TOF, MOF or BOF, or cold, warm, hot	The stage picks the framework, the structure and the CTA. Without it the skill has to guess and you get copy aimed at the wrong temperature	Required
The right workspace active: `_default` for your agency, `client--*` for a client	Agency mode writes ads that sell your agency, client mode writes ads that sell your client's business. The wrong workspace produces the wrong audience entirely	Required
`brand/[workspace]/verified-claims.md` with entries marked `consent: ok-as-stated`	Every number, result, testimonial and named client in an ad has to trace to it, or it gets genericised and flagged as a proof gap	Required
`epistemology.md` in the workspace	Ad hooks are built from your invisible assumptions and reframes. Without it the skill warns you and writes benefit-led ads instead, and the skill itself says to expect lower click-through	Optional
`ica-output.md` and `offer-output.md`	The ICA supplies audience pain, desire and language for the hooks. The offer supplies the value stack and pricing for BOF ads	Optional
`lead-magnet-output.md` if the ad drives to a lead magnet	The skill pulls the magnet's title, hook and value proposition straight into the TOF and MOF copy	Optional
A GHL sub-account for the client, if these are client ads	Client ad copy is formatted for AI Ads inside that sub-account. No sub-account, no deployment path for the copy	Optional

How it runs

Fix the funnel stage and the ownerIt confirms whether this is TOF, MOF or BOF, and whether the ad sells your agency or a client's business. It asks for your product or service, audience, campaign goal, budget level and any existing performance data.
Select a hook formulaIt chooses from eight formulas based on intent and audience temperature: big claim, question, if-then, secret, us versus them, pattern interrupt, call out, controversial. The hook is drawn from an assumption-challenge or reframe in your epistemology file, not from a generic benefit.
Apply the body structure for the stageTOF gets PAS, story or educational. MOF gets case study or webinar and VSL structure. BOF gets an offer stack or urgency and scarcity build. Each structure is a fixed beat sequence the copy fills.
Add the CTA and check the limitsThe CTA button is matched to commitment level: Learn More at TOF, Register Free or Book a Call at MOF, Get Started or Claim Offer at BOF. It then checks primary text against 125 visible characters, headline against 40 and description against 30, and front-loads the hook above the fold.
Apply an expert lensHigh-ticket offers get the Hormozi value equation, nurture gets Kern's value-first then retarget sequence, attention plays get Billy Gene's pattern interrupts, low budgets get Portie's five-dollar approach.
Run the diagnosis gateIt scores Attention, Resonance, Belief and Action 1 to 5 in reading order. The first axis below 4 is the diagnosis and gets fixed first. All four must reach 4 before the copy is considered deployable, and a model grader runs over customer-facing ad sets and blocks on any failure.
Run the claim checks and format for deploymentIt lints the copy against your verified-claims registry and spawns a claim grader before presenting. Then it formats: agency ads for Meta Ads Manager, client ads for GHL AI Ads input with targeting recommendations attached.

What you get

Primary text, headline and description written to Meta's visible character limits
The CTA button selection matched to funnel stage
The hook formula and body structure used, named
Copy formatted for its deployment path: Meta Ads Manager for agency ads, GHL AI Ads input for client ads
Targeting recommendations you can set in GHL, for client ads
A diagnosis score across Attention, Resonance, Belief and Action with the first breakdown named
An entry appended to brand/[workspace]/assets.md after an ad copy set of three or more variations

Honest limits

Read this before you rely on it

Specifics are capped by your registry. A case study ad without a client result in verified-claims.md comes back generic with [PROOF GAP: ask the client for a real number/result here] where the number should be. The four agency templates that lean hardest on proof, case study and authority ads, are the ones that suffer most from an empty registry. Fill it before you write those.
The claim checks do not see chat. The linter fires on file writes to client-facing deliverables, so ad copy you copy straight out of the conversation into Ads Manager has passed only the model grader, not the file hook. Read the proof gap flags before pasting.
Missing epistemology.md costs performance, not just depth. The skill says so directly: it falls back to benefit-led hooks and you should expect lower click-through.
It writes text only, and it does not publish. No image, no video, no thumbnail, and it does not create the campaign, the ad set, the audience, the budget or the Lead Form. Agency copy still has to be loaded into Meta Ads Manager by you, and client copy into AI Ads inside the client's GHL sub-account. Copy without a creative and a campaign is not a running ad.
No performance prediction. The diagnosis score is a rubric applied to the draft, not data. Which hook actually wins is decided by spend.

Where people go wrong

The mistake	Do this instead
Asking for 'some Facebook ads' with no funnel stage	Say TOF, MOF or BOF, or cold, warm, hot. The stage decides the framework, the structure and the CTA button. Everything downstream is wrong if this is wrong.
Running the skill in the wrong workspace	Check `brand/_active.md` first. In `_default` it writes ads selling your agency. In a `client--*` workspace it writes ads selling the client's business to their customers.
Expecting it to build the campaign as well as the copy	Campaign structure, audiences, budgets and Lead Forms are a different job. This skill hands you the four text fields and the CTA choice.
Writing headlines that read well but get truncated	Keep the headline to 40 characters and the description to 30, and put the whole hook inside the first 125 characters of the primary text. Everything past that is below the fold.
Asking for a case study ad before the client has attested to the result	Get the number and the consent into `verified-claims.md` first. Otherwise the ad's central claim comes back as a proof gap flag.
Requesting one ad and shipping it	Ask for a set. Three or more variations gives you something to test and triggers the learning loop that tunes future hook style and CTA wording to your feedback.

Worth knowing

The hook is the whole ad, and the skill will only produce a strong one if epistemology.md gives it something to argue against. The pattern that converts cold traffic is 'you believe X, the deeper truth is Y', which needs a real invisible assumption from your audience on file. Run the profile-deepening work once and every ad set after it gets better hooks, rather than benefit statements dressed up as hooks.