Agent-Skills.md

Agent Skills: Content Distribution

Turn one real idea, vault page, article, photo, screenshot, code diff, spec excerpt, or diagram into platform-native content for LinkedIn, X, Reddit, TikTok, Instagram Reels, YouTube Shorts, Medium, Substack, or a personal-site article. Use when creating or editing files in variants/, choosing a channel or platform for a page, republishing or updating content that already shipped, or figuring out what's currently live for a given page.

UncategorizedID: epicenterhq/epicenter/content-distribution

Repository

epicenterhq/epicenter
EpicenterHQLicense: NOASSERTION
4,687363

Install this agent skill to your local

pnpm dlx add-skill https://github.com/EpicenterHQ/epicenter/tree/HEAD/.agents/skills/content-distribution

Skill Files

Browse the full folder contents for content-distribution.

Download Skill

Loading file tree…

.agents/skills/content-distribution/SKILL.md

Skill Metadata

Name
content-distribution
Description
Turn one real idea, vault page, article, photo, screenshot, code diff, spec excerpt, or diagram into platform-native content for LinkedIn, X, Reddit, TikTok, Instagram Reels, YouTube Shorts, Medium, Substack, or a personal-site article. Use when creating or editing files in variants/, choosing a channel or platform for a page, republishing or updating content that already shipped, or figuring out what's currently live for a given page.

Content Distribution

Follow writing-voice for tone. Use social-media when drafting final LinkedIn, X, or Reddit post copy.

Core Philosophy

The goal is not to become a full-time creator. The goal is to make existing thinking travel farther.

Use one markdown source. Use real artifacts. AI adapts, packages, resizes, rewrites, captions, and formats. AI does not pretend to be the author.

real idea
  -> pages/<slug>.md (the private draft)
  -> variants/<slug>-<channel>-<format>-<version>.md (the shaped artifact)
       placements: [{platform, published_at, url, caption}] (shipped facts)
  -> performance notes
  -> next ideas from replies

Default Workflow

  1. Identify the source artifact: a pages/ note, article draft, photo, screenshot, code diff, ASCII diagram, spec excerpt, voice note, or product decision.
  2. Distill one content atom: thesis, tension, proof, visual, audience, and desired reaction.
  3. Choose renderers by platform, not by rewriting the idea from scratch.
  4. Preserve the human thesis and concrete examples. Let AI adapt structure and phrasing.
  5. Produce wrappers for each platform: hook, caption, title, CTA, and format.
  6. Keep performance notes simple: hook, visual type, platform, replies, saves, shares, profile clicks, and next variant.

Content Atom

Before rendering, reduce the source to this shape:

Thesis:
  The claim the post is making.

Tension:
  Why the claim matters or what common belief it pushes against.

Proof:
  Concrete artifact, example, diff, screenshot, metric, failure, or quote.

Visual:
  Real photo, screenshot, diagram, code, spec excerpt, or Marp slide.

Audience:
  Who should feel seen, challenged, or helped.

Desired reaction:
  save, argue, try, reply, share, click, subscribe.

Renderer Decision Tree

Does the user need platform versions?
  -> Use references/platform-renderers.md.

Does the output need strong opening lines?
  -> Use references/hooks.md.

Does the output include carousels or short video?
  -> Use references/marp-remotion-pipeline.md.

Platform Grouping

Treat TikTok, Instagram Reels, and YouTube Shorts as one short-video renderer with small platform wrappers:

Same:
  core idea, slides, photos, screenshots, voiceover, captions.

Different:
  first hook frame, title, caption, CTA, pacing if needed.

Treat Medium and Substack as one article renderer with different relationship posture:

Medium:
  discovery and searchable article.

Substack:
  relationship, continuity, and personal context.

Treat LinkedIn and X as related but not identical:

LinkedIn:
  canonical concise public argument with one strong visual.

X:
  fragments, threads, sharper hooks, higher frequency.

Treat Reddit separately:

Reddit:
  native subreddit post or comment. Rewrite around the community. Do not dump recycled promo.

Source Of Truth

pages/<slug>.md is the source. channel is the content identity (bradencodes, braden-essays, epicenter, ...). In the sibling vault repo, channel promises live at ../vault/specs/20260609T010000-channel-promise-approval-ledger.md; read that file before picking a channel for a page for the first time because each channel's "what belongs" and "what does not belong" sections are normative. format is article | social-post | short-video. A social-post body is one or more short written segments separated by ---: one segment is a one-shot post, several are a thread. platform is the distribution surface, and each format ships only to its platform class (the lint enforces this matrix):

article      → personal-blog | medium | substack | newsletter
social-post  → x | linkedin | reddit
short-video  → instagram | tiktok | youtube

A variant's filename is <page>-<channel>-<format>-<version> where version is the YYYY-MM-DD checkpoint the variant was cut from the page:

pages/2026-06-15-my-page.md
  -> variants/2026-06-15-my-page-bradencodes-social-post-2026-06-20.md
       placements: [{platform: x, published_at, url}]
  -> variants/2026-06-15-my-page-braden-essays-article-2026-06-20.md
       placements: [{platform: personal-blog, published_at, url}]

Freeze-on-supersede

The latest variant for a (page, channel, format) lineage is living: edit it in place to track the page. Cutting a new version freezes its predecessor — never touch a superseded variant again; it is the curated record of what shipped.

  • Content changed enough to preserve the old checkpoint? Cut a new variant with today's date as its version; the old one freezes.
  • Shipping the living variant somewhere (first ship, update, or repost)? Append a placement to its placements[] — one object per platform shipment.

What's currently live is derived, not stored: for a given platform, the current placement is the one with the latest published_at on the latest variant that carries that platform. Older variants and placements are history, not stale data to clean up.

Current vault schema: pages/matter.json requires title, date, timezone, and status; variants/matter.json requires page, version, channel, and format, with optional subtitle (a channel-specific dek; the page's own title stays private and canonical). placements[] is validated by bun run publishing-lint, not Matter: each object has platform plus optional published_at, url, and caption. Leave published_at off until content is actually live.

Do Not

  • Do not generate pure AI images or pure AI videos when the user asked for authentic content from existing materials.
  • Do not turn every platform into a separate original writing task.
  • Do not optimize for raw volume before a platform wrapper preserves the author's taste.
  • Do not add fake vulnerability, fake lessons, fake metrics, or invented personal stories.
  • Do not let platform advice override the source thesis.

Vault References

  • ../vault/specs/20260609T010000-channel-promise-approval-ledger.md: channel promises and what belongs on each channel.
  • ../vault/specs/20260611T230004-channel-routing-cheat-sheet.md: quick channel routing decisions.
  • ../vault/specs/20260613T181206-channel-routing-test-batch.md: examples that test the routing rules against real pages.