One Source Into Thirty Assets . Agency OS

What it does

Breaks one piece of pillar content into standalone assets that never reference the source: LinkedIn and Instagram carousels, X threads, Facebook comment-threads, a LinkedIn newsletter, short-form video scripts, image posts per platform, thumbnail briefs, and a blog post if the source was video or audio. Seven modes cover everything from full atomisation to a single-platform batch to a hook-only run. Every asset is voice-matched to the workspace, tied to a positioning pillar, and given a platform-appropriate CTA. Deliverables are saved as .md and .docx in your workspace content folder.

Say this to start

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

/repurpose

> repurpose this

> atomize this

> turn this into everything

> make a thread

> full content system

When to reach for it

you just recorded a webinar, podcast or coaching call
one blog post needs to become a month of social
you need a carousel or a thread from something you already made
you have a backlog of content and want to know what to repurpose first
you need a posting schedule for assets you already have
you want to combine several short posts back into one pillar piece

When NOT to use it

If you actually want	Use this instead
cutting and rendering the actual video clips from a recording	`video-repurposer`
the edit itself, silence removal and captions	`auto-editor`
planning an ongoing calendar and pillars with no source asset to atomise	`content-system-architect`
a standalone long-form article written from scratch	`blog-writer`
a single caption written from a transcript	`social-caption-writer`
image or video generation prompts as the deliverable	`creative-director`

Before you start

What you need	Why
The source asset itself, pasted, as a file path, or as a URL	there is nothing to atomise without it, and the skill will not invent a source	Required
A source that clears the minimum length for the mode	Mode G needs 10 minutes or 2,000 words of transcript, 1,000 words of blog, or 3 minutes of scripted video; below that it sends you to Mode B instead	Required
Brand context from Build 1 (Brand DNA), or the answers by hand	without a workspace it asks you for business name, product, audience, brand voice, differentiator and social proof before it will write anything	Required
A decision on which product the repurpose is for, in a multi-product workspace	it asks, because the output folder and the CTA alignment differ per product	Required
`verified-claims.md` with consent-tagged proof	stats, testimonials and results that are not registered get genericised or flagged [PROOF GAP], so you supply the proof or you publish without the number	Optional
A Canva MCP connection, or Playwright, or the workspace's image tool	these are what turn carousel copy into actual designed slides; with none of them you get the copy plus design specs and you build the visuals yourself	Optional
`ica-output.md`, `offer-output.md`, `voice-profile.md` and `content-plan.md`	all optional, but they are what stop the assets sounding like generic marketing and keep the CTAs pointed at your real offer	Optional

How it runs

Load the workspaceDetects whether you are in agency mode or a client workspace, loads brand memory under freshness rules (files older than 90 days are skipped with a suggestion to refresh), and reads past feedback tagged to this skill. A pattern seen three or more times is treated as a rule.
Find the angle before atomisingRuns the seven angle lenses on the source's core idea and picks the sharpest non-consensus cut. This happens first on purpose: a repurpose that just shortens the original inherits the original's framing.
Pick a modeSeven modes: A full atomisation, B single platform, C content audit with impact and effort scoring, D distribution calendar, E hook generation, F reverse atomisation into a pillar piece, G everything with visuals. It will infer the mode from how you ask, or you can name one.
Analyse the source and extract atomic unitsPulls the standalone ideas, frameworks, stories and lines out of the source. These units are the shared input for everything downstream, so this step runs before any writing.
Research each topic per platformIn Mode G this is mandatory before writing: the chosen topic for each format is researched on its target platform first, so the thread is written to what works on X and the carousel to what works on LinkedIn.
Transform through the recipesTwenty-six named transformation recipes map source material to platform formats. Each asset gets its own voice adaptation, its own hook, and a CTA appropriate to its platform. Formats are generated in parallel, one whole format per batch.
Generate visualsChecks what visual tooling is connected and steps down through the options: a design MCP for the highest-quality carousels, then browser-rendered HTML templates, then the workspace's image tool for key visuals, then copy plus design specs for manual creation.
Quality gate and learning loopEvery hook in the batch goes to a hook grader and weak ones are rewritten. Assets carrying a direct CTA or offer also go through a copy grader. The critique skill scores the top three assets and flags anything under 70. Then it asks how the output landed and logs your answer.

What you get

All deliverables as .md and .docx in [target]/content/repurposed/, named [YYYY-MM-DD]-[source-topic-slug]-[mode]
Mode G produces a 13-item set: LinkedIn carousel (10 slides), Instagram carousel (10 slides), an 8 to 12 tweet thread, a Facebook comment-thread post with 5 to 10 comments, an 800 to 1,500 word LinkedIn newsletter, a 60-second video script, platform captions, a 1,500 to 3,000 word blog post if the source was video or audio, 4 to 8 image posts each for Facebook, Instagram and LinkedIn, and vertical plus long-form thumbnail briefs
Mode A produces 15 to 30-plus assets across platforms with a staggered posting sequence
Mode C produces an impact and effort scored audit of your existing content, with no writing
Mode D produces a 7 to 30 day distribution calendar
Mode E produces 5 to 10 scored hooks per platform
Critique scores on the top three assets, with anything below 70 flagged for revision

Honest limits

Read this before you rely on it

It produces copy, briefs and designed stills. It does not cut, caption or render video files. If you want the actual clips out of a recording, that is video-repurposer and auto-editor.
Without a design MCP, browser automation, or an image tool connected, you get copy plus design specs. You are then building the carousels and image posts yourself.
Output quality is capped by the source. The skill's own rule is not to atomise weak source material: if the pillar piece has no clear insight, thirty derivatives of nothing is still nothing.
The quality gate flags assets scoring below 70 and suggests revision. It does not silently fix them, so that judgement call comes back to you.
Stats, testimonials and social proof are only used if they already exist in your workspace files or in what you supplied. Anything else is genericised or marked as a proof gap for you to fill.

Where people go wrong

The mistake	Do this instead
Feeding it a thin or rambling source and expecting 30 strong assets	Check the source against the minimum length table first, and improve the pillar piece before atomising it. Ten strong pieces beat thirty mediocre ones.
Publishing the whole batch on the same day	Stagger across 7 to 14 days. Mode D exists to build that schedule for you.
Pasting the same text to every platform	Let it produce a separate voice adaptation per platform. Each asset is written to that platform's specs, and wrong lengths or dimensions cut reach.
Leaving 'as I mentioned in the video' style references in derivatives	Every asset has to stand alone with zero references to the source. If a line only makes sense to someone who saw the original, cut it.
Asking for full atomisation when you only need one platform	Say which platform. Mode B produces 3 to 10 assets for a single platform in a fraction of the time.

Worth knowing

Run Mode C first on a backlog you have not touched. It scores existing content by impact and effort without writing anything, so you find out which one asset is worth the full atomisation before you spend the day on the wrong one.