make-content

What you get

One of five things, saved to the brand's workspace and shown in chat:

1. A repurposed set. One finished asset is mined into scored units. Each platform the brand uses gets a native version with a different lead unit and hook, ready to paste. Each version has an asset line saying what visual it needs and who makes it.
2. A single post. One LinkedIn, X, Threads or Instagram post, a thread, or an X long post or Article (references/formats.md), written native to its platform. It skips the calendar and the units table unless there is a source asset.
3. A fresh piece. A one-page brief, then a first draft in the brand's (or a named author's) voice. Long-form adds the publish kit: headline options, meta, slug, H2 skeleton, internal links, CTA, image notes and distribution notes.
4. A refresh redline. An old piece rewritten where it decays, with each change and its reason.
5. This week's newsletter issue, written from a saved template and the week's notes.

Every piece carries an edit note and a claims check: what was kept, softened and cut, and the proof that would restore each cut. The formats are listed in references/formats.md. Anything else is routed (Out of scope).

Before you start

Read research/market/positioning-<product>.md when present, else research/market/positioning.md, and its dated report. Carry the owner-selected launch direction into this work even when the live site/kit still uses earlier messaging. Unselected options stay proposals and every factual claim keeps its existing proof/approval status.

Briefs only, one or several topics: after the shared setup and evidence read below, run the existing make-content-content-brief skill (../make-content-content-brief/SKILL.md) for each supplied topic. Reuse a shared inventory/source read, distinguish overlapping intents, and deliver the requested briefs with their sources and proof gaps. Stop at briefs: no draft articles, outline-approval loop or newsletter intake unless asked. State any unfinished topic; never return one brief as a completed ten-topic request.

1. Setup first: do the marketing-guide skill (../marketing-guide/SKILL.md) §1–2 (connect, brand, brain read) before anything else; the job line is "I write from your brand's voice and claims through it." (get_started with wants: "more".) Then, for this playbook:

   • Brain path: open chapters 2 "What we sell", 5 "Who we sell to", 11 "How we sound" and 22 "Everything we have made and run" (past content), then gooseworks:brand_read lean with only sections: ["sources","organic"] (the brain doesn't hold own posts or saved sources).
   • Fallback path: gooseworks:brand_read lean with sections: ["summary","kit","learnings","sources","organic","products"], then product_ids for the products the piece covers. Learnings at detail: "full" only in their own call, when you need do, insight or template_hint rules (several sections at full detail overflow). gooseworks:knowledge_search with the task, for past content and documents.
   • Either path: gooseworks:file_list on content/ and brand/voices/; gooseworks:file_read content/channels.md.
   • The brand's platforms, in this order: content/channels.md → social/accounts.md → sources rows with relationship: "self" and source_type: "organic_social", plus organic.own_accounts. Ignore the ad-library sources.

2. Say plainly what the brain can't support, before drafting. Lines this piece needs (the core positioning, a number, a customer name) that aren't signed off go to the marketer in the intake message, per Approval gates § Claims. When something the marketer says contradicts the brain, ask which is true and save the answer as their correction.

3. Ask only what is missing, in one intake message, at most four questions.

   • Repurpose: the source asset, and the platforms if not saved.
   • Single post: the platform, the point and whose account it is for (brand page or a named person), if not given.
   • Fresh piece: topic, audience, format, where it publishes and by when, and the CTA destination.
   • Opinion-led piece: whether the marketer has a view.
   • Case study: the interview transcript, the customer's written OK to be named and quoted, and the results period.
   • Search-intent piece: the yes for the search-results read when it needs the proxy (Steps, 4.1).

   Never guess any of these. With no topic, offer 3 candidate questions from the voice-of-customer bank, labelled "unverified: you pick". Save the platform list and house style to content/channels.md; they are not learnings.

Steps

1. Pick the path. Repurpose · single post · fresh piece · refresh · newsletter.

• A sibling's request arrives as a page brief (search/briefs/<page-slug>.md), an asset-list row (gooseworks:campaign_read) or a "what to make next" line (reports/performance/<date>.md). Treat it as a fresh piece with that input.
• Read any Graphic ready: answers under your earlier asset lines and image notes (content/repurposed/, content/drafts/). Attach those images to the pieces they belong to.
• Do only the work the ask needs.

2. Repurpose.

• Get the source in (references/repurposing.md §1).
• Mine and score units (the make-content-source-to-units skill, ../make-content-source-to-units/SKILL.md). A thin source, such as a short silent clip, supports 2–3 versions. Say so, and suggest a richer asset.
• References only, never to copy: gooseworks:social_inspiration_search, or brand_read organic.competitor_top_posts (skip posts outside the brand's category).
• Write one native version per platform the brand uses, each led by a different unit (references/repurposing.md §4). Skip the platform the source already lives on unless a re-cut is asked for. Add a second surface on a platform only when strong units outnumber platforms.

3. Single post.

• Write it to references/repurposing.md §4 for its platform and account (brand page or a named person), with the hook first and 2 alternatives.
• Voice: the account's card if one exists; else the brand card (build it free from brand_read organic when ≥8 own posts exist); else match against the make-content-two-pass-edit skill (../make-content-two-pass-edit/SKILL.md)'s fallback order and label the post "voice provisional". Never open a paid read or the voice interview for one post; offer to build a card afterwards.
• Then step 6. No brief, no calendar intake.
• If the marketer wants a run of posts over dates, say you can plan a two-week calendar instead, and stop here.

4. Fresh piece. In this order:

1. Brief (the make-content-content-brief skill, ../make-content-content-brief/SKILL.md).
   • Search-intent pieces (SEO, answer, comparison, a how-to people search for), when no page brief carries the top results: read them first. Use the host's own web search (free), or one /v1/google/search read through the proxy (the marketing-data-access skill (../marketing-data-access/SKILL.md), the spend-gate skill (../spend-gate/SKILL.md) proxy branch, with the yes asked in the intake message).
   • Open the top 3 results and log URLs and dates. With no fetch, judge information gain from titles and snippets and label it "snippets only". "Information gain: not checked" ships only if the marketer declines the read.
   • Then the public sources for world facts.
2. POV. If a view is needed and missing, run the interview (references/point-of-view.md). With no answer, write the no-POV version and label it.
3. Proof gate. Count only H2s that make factual claims; interview lines count as proof. If fewer than half are proofed, stop at the brief and hand over "what unlocks this draft" (4–6 specific asks).
4. Outline gate. Long-form and every case study: show the brief and outline and wait for a yes.
5. Draft.
   • Bylined pieces: the author’s card through the voice-card atom. For company blog/article work, read brand/voices/blog-<brand-slug>.md first when present; apply its company style and bans, then a named author’s card for individual differences. A provisional guide stays labelled provisional. Social cuts use its applicable channel rules, not blog length.
   • Draft to the format's anatomy (references/formats.md), section by section for long pieces, saving each with gooseworks:file_write.
   • Fill proof only from the brief's inventory; each gap stays [PROOF NEEDED: …]. Before leaving any fill-in or gap, check the brand's own public docs (help centre, quickstart, pricing, changelog): a product fact found there (e.g. what one video costs in credits) is filled in with its link, not asked. Fill-ins should shrink between drafts, never grow.
   • A page draft puts its search/briefs/… path on line one.

5a. Refresh.

• Get the current text in.
• Keep what earns its place. Rewrite the lede and title.
• Update stale facts only to sourced or signed-off ones. Add what the decay reason points to, and cut what's outdated.
• Deliver a section-by-section redline: old → new, and why.

5b. Newsletter.

• If content/newsletter/template.md is missing, send one message with:
  • a proposed default: a 300–500-word editor's letter, 2–4 items each with "why it matters", one CTA, a sign-off;
  • at most four questions: who signs it, who the list is and how they joined, send day, CTA destination;
  • the ask for this week's notes.
• With no notes, offer an issue built from the brain, labelled "unverified: you pick".
• Save the template. On later runs, ask only for the week's notes.
• The lead is an editor's note from the notes, in the signer's voice. It needs the POV interview only when it argues a position.

6. Edit, check, edit (the make-content-two-pass-edit skill, ../make-content-two-pass-edit/SKILL.md):

1. The substance pass.
2. The claims check, per Approval gates § Claims. When a piece is sponsored, affiliate, gifted or a creator collaboration, also load ../ad-policy-check/disclosure.md (in the ad-policy-check skill) and apply it.
3. The readability pass and the AI-tell check on what survived.

Attach the edit note and one claims block per piece or set.

7. Hand over.

• Show, in order: title and path; the brief or a link to it; the piece; the publish kit (long-form); the edit note; the claims check; open gaps ([PROOF NEEDED], [CUSTOMER APPROVAL NEEDED], [NAME: …], PERSON YES NEEDED).
• For a large set, show the strongest three in full and say the rest are in the file.
• End with: "Nothing is published — you edit and post. If you want the social pieces in a two-week posting calendar, ask for it next." Offer only; never start another playbook yourself.

Approval gates

• Claims. Brand claims go only through the shared claims gate (the claims-gate skill, ../claims-gate/SKILL.md): signed-off claims verbatim, or its scoped sign-off now (3–5 lines put to the marketer in the intake message, their answer recorded; never when someone else approves claims — then those lines wait), or plain product facts only, which is also the path when the marketer declines. World facts, competitor facts, stories, sensitive topics and third-party rights follow references/fact-rules.md. Never self-approve. A case study with unapproved names or numbers is delivered as blocked.
• Point of view. Opinions, stories and first-person experience come only from the marketer or the named person. If missing, ask. If they decline, use the no-POV format.
• Bylines and named people. Anything published under, or quoting, a named person carries PERSON YES NEEDED: that person's own yes.
• Outline. Long-form and case studies: the brief and outline get a yes before drafting.
• Publishing. Nothing goes live from here.
• Money. The default path is free. A proxy read (top search results, post history) or any import that quotes a price follows the spend gate: a probe with a yes, then a ceiling.

Output and write-back

• Workspace (brand coworker scope; <D> is YYYY-MM-DD):
  • content/briefs/; content/drafts/<D>-<slug>.md (single posts and fresh pieces); content/repurposed/<source-slug>/<D>.md; content/refresh/<D>-<slug>.md
  • content/newsletter/template.md and content/newsletter/<D>.md
  • content/pov/ (interview answers, reused by the social calendar)
  • content/channels.md (platforms and house style)
  • voice cards at brand/voices/<handle>.md and brand/voices/blog-<brand-slug>.md, through the voice-card atom
• Brand documents via the shared save-to-brain atom (../save-to-brain/SKILL.md):
  • The newsletter template, versioned: supersede, so the previous version is retired.
  • Drafts are never imported; they aren't brand facts.
  • When the marketer says a piece is published at a public URL, import it as content inventory (gooseworks:knowledge_import url, then gooseworks:job_get kind: "knowledge_ingest"). If the answer quotes a price, stop and ask.
• Learnings only via the shared save-learning atom (../save-learning/SKILL.md). Write only writing rules the marketer stated or confirmed in their own words: prefix [content YYYY-MM-DD], one idea each, superseded own learnings retired. An inferred rule is asked as a question first.
• Not yet stored in the brain: a content library, a story bank. The workspace files are the interim store.

Done when

Every piece is saved and shown with, per path:

• Repurposed set: its units table, and one native version per platform.
• Single post: the hook and two alternatives.
• Fresh piece: its brief; for long-form and case studies, the marketer's outline yes recorded before the draft; the publish kit for long-form.
• Refresh: the section-by-section redline with a reason per change.
• Newsletter: the saved template.

And on every path: its edit note; one claims block; every unfilled gap tagged [PROOF NEEDED: …] (or [CUSTOMER APPROVAL NEEDED], [NAME: …], PERSON YES NEEDED). Nothing was invented.

Out of scope

• A dated posting calendar, keeping accounts alive, posting: run-organic-social, which slots these cuts.
• Ad copy and creative: make-static-ads (image ads); video ads: the goose-video skill.
• Rendering images or carousels: make-static-ads; produced video: the goose-video skill. Here: words, scripts, slide copy, shot notes.
• Keyword research, content audits, visibility checks: improve-search-visibility, which sends page briefs here.
• Battlecards — research-competitors, which uses the saved rival evidence.
• Which topics get the month's capacity: the marketer, with plan-campaign's brief.
• Promo email broadcasts and flows: email-lifecycle. The editorial newsletter issue is here.
• Approving claims: the scoped sign-off runs inline through the claims gate. A full claims audit and signing off the brand voice are not covered yet.
• Landing pages, product and listing pages, press releases, sales decks, one-pagers, sponsor reads, translation: not covered yet. Say so plainly, without naming internal plans.