Direct Response Copy Engine . Agency OS

What it does

Produces finished persuasive copy: email sequences, landing and sales pages, sales letters, webinar scripts, organic social posts, headlines, hooks and subject lines. In agency mode it also writes cold outreach, proposal sections, case studies, retainer pitches, partnership outreach and testimonial requests. Every piece comes back with the framework used, the awareness level targeted, a quality score and two to three alternatives to test. It is built on 12 master copywriters' frameworks and refuses to hand over copy that fails its own gates.

Say this to start

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

/write-copy

> write copy

> email sequence

> landing page

> improve my copy

> copy audit

When to reach for it

Writing an email sequence or a single email
Writing a landing page or long-form sales page
Writing headlines, hooks or subject lines to test
Auditing or rescoring copy you already have
Agency outreach: cold email, proposal copy, retainer pitch
Organic social posts written from scratch, not from a transcript

When NOT to use it

If you actually want	Use this instead
Meta, Facebook or Instagram ad copy with funnel-stage structure and platform character limits	`meta-ads-copywriter`
A ready-to-record video sales letter with timing markers and a production package	`vsl-architect`
Platform captions generated from a video transcript	`social-caption-writer`
Extracting or defining the brand voice itself	`brand-dna`
Long-form SEO blog articles	`blog-writer`
The visual or image direction behind an ad	`creative-director`
Building the funnel pages and CRM sequences that host the copy	`funnel-architect`

Before you start

What you need	Why
An active workspace with `voice-profile.md`	Without it the copy is written in a neutral tone rather than yours	Required
`brand/[workspace]/verified-claims.md` with entries marked `consent: ok-as-stated`	Any number, result, testimonial or named client that is not in the registry gets genericised and flagged as a proof gap instead of written	Required
`epistemology.md` in the workspace	Hooks, headlines and angles are drawn from your invisible assumptions and reframes. Without it the skill warns you and writes from the owner profile alone, and the output is shallower	Optional
`policy.md` with your never_touch list	Those topics are treated as a hard deny across every copy variant	Optional
Real customer language: reviews, testimonials, call notes or survey answers	Step 0 mines your buyers' exact pain, desire and objection phrases. Without them the skill reasons about the customer instead of quoting them	Optional
`ica-output.md` and `offer-output.md` if they exist	They give the skill structured audience and offer context rather than making it ask you	Optional

How it runs

Mine voice of customerBefore writing anything it extracts 3 to 5 pain phrases, 3 to 5 desire phrases and 2 to 3 objection phrases in your buyers' own words. Give it reviews, testimonials or call transcripts here if you have them.
Set copy type and awareness levelIt asks what you are writing and loads the matching reference. Then it places your reader on the five Schwartz awareness levels, from unaware to most aware, because that decides what the copy leads with.
Select a frameworkCold audiences get AIDA or Hook-Story-Offer, problem-aware get PAS, transformation stories get Before-After-Bridge, B2B gets the 4Ps, long-form sales gets PASTOR. It names the framework it picked and why.
Write the pieceHook, body and close, with proof stacked as specifics rather than generalities and one single CTA. It resolves CTA mode first: direct, soft or none. A mode you set is not its to change, and under none the CTA slot simply stays empty.
Apply conversion optimisationIt runs the value equation, anchoring, loss aversion and social proof over the draft, checks you are asking for the next step rather than a three-rung skip, and strips friction.
Run the quality gatesGate 0 names the promise and the pain in one sentence each. Gate 0.5 scores Attention, Resonance, Belief and Action 1 to 5 in reading order and the first axis below 4 is the diagnosis. Then a red-flag scan, section scoring, and the Ogilvy 100-point evaluation against a per-format minimum.
Run the claim checksBefore it shows you the final draft it lints the copy against your verified-claims registry and spawns a model grader over the text. If either finds an ungrounded specific, it genericises or grounds it and re-runs rather than presenting the draft.
Deliver and logYou get the copy plus framework, awareness level, scores and alternatives. After major pieces it appends the deliverable to assets.md and asks you how it landed so the feedback lands in learnings.md.

What you get

The complete copy, formatted for its platform
The framework used and why that master's approach was chosen
The awareness level targeted and how the copy matches it
A quality score: four diagnosis axes 1 to 5 with the first breakdown named, plus the Ogilvy 100-point score with breakdown
Two to three alternative headlines, hooks or subject lines, each testing a different angle rather than different wording
An A/B test priority: what to test first and the expected impact
An entry appended to brand/[workspace]/assets.md and feedback logged to learnings.md

Honest limits

Read this before you rely on it

Specifics are capped by your registry. Numbers, percentages, dollar figures, named results, testimonials, credentials, named people, dated events and first-person anecdotes only appear if they trace to brand/[workspace]/verified-claims.md with consent: ok-as-stated. Everything else comes back in generic framing with [PROOF GAP: ask the client for a real number/result here] in the copy. That is deliberate, and it is why the copy will feel thinner until you fill the registry.
The claim checks are a backstop, not a full net. The linter and the hook fire on file writes to client-facing deliverables. They do not see text the model prints in chat that you paste straight into Canva, a CRM or an ad manager. Prevention is the real defence, so read the flags before you copy anything out.
The Ogilvy score and the diagnosis axes are the skill's own rubric applied to the draft, not market performance data. Real conversion still has to be tested, which is why it hands you an A/B priority.
Missing epistemology.md costs you the good hooks. The skill warns you and continues from the owner profile only, and it flags copy that does not draw on at least one epistemology pattern as generic by default.
It writes words, not pages or images. Building the page, the sequence and the CRM plumbing is a separate job, as is the visual that sits behind an ad.

Where people go wrong

The mistake	Do this instead
Asking for copy without saying whether you want a CTA	State the CTA mode up front: direct, soft or none. A mode you set is locked, and under `none` nothing fills the empty slot, not a link, not a follow prompt, not a closing question.
Letting the skill write 'link in bio' or a specific URL	Only confirm that language once the link is actually live. The skill will not assert something about your profile it cannot see, and you should not either.
Stacking several asks into one piece: book a call, download this, follow us	One primary CTA per piece. Everything else is a distraction that costs conversions.
Polishing the CTA when the piece is not converting	Score the four axes in reading order and fix the first one below 4. If the opening loses the reader, nothing later in the piece matters.
Expecting hard numbers and named case studies with an empty registry	Fill `verified-claims.md` first with entries the client has attested and consented to. Every proof gap flag in the draft is a question to take back to them.
Writing to everyone at once	Name one reader, their role, their pain and their world. Copy aimed at all audiences sounds generic to every audience.

Worth knowing

Paste real customer language before you ask for anything: reviews, support tickets, call transcripts, survey answers. Step 0 mines your buyers' exact pain, desire and objection phrases and puts them straight into the copy, which is what makes a reader feel it was written about them. Without that raw material the skill still writes well, it just writes about your customer instead of in their words.