Content Repurposing Line with a Human Approval Gate

You record one good thing a week. The clips, posts, captions and newsletter never get made, because each one is 20 minutes of unrewarding work.

The tools that promise to fix this auto-post — which is how a brand account ends up with something nobody wanted on it.

Every node with its icon, as on the n8n canvasstarts a runa model or a parser, wired into the node above itHover a node for its name and type

• 01Ingest & Transcribe

  20 nodes · 24 connections · 5 notes

  form or manual in, transcribe, one canonical record

• 02Segment & Rank

  11 nodes · 10 connections · 3 notes

  windows cut, the model picks IDs, anchored, top five

• 03Draft & Rules Check

  13 nodes · 12 connections · 3 notes

  the brand rules in, drafts out, a check in code

• 04Approval Queue

  9 nodes · 7 connections · 2 notes

  the review card, and a nudge at 48 hours

• 05Approve / Reject Callback

  17 nodes · 16 connections · 3 notes

  the signed callback: publish, or learn

The gates, 11 If and Switch nodes, decide whether anything happens next: a countable part of the system, not a claim about it. The 16 sticky notes carry the reasoning on the canvas, at the point it applies.

HoverTap a card to read what it prevents.

• One path to publishing

  One path to publishing

  A backwards walk from the publish node returns exactly one path, and every step of it is a gate. That is a property of the graph, not a promise about it.

• The model never writes a timestamp

  The model never writes a timestamp

  It cannot count seconds. It picks a window already cut from the transcript, and the clip’s text is the window’s own words, not a paraphrase.

• The rules check is code

  The rules check is code

  A prompt asking the model not to use a banned word is a request. Banned words match on word boundaries — “cure” does not fire on “accurate” — and one blocked draft blocks the whole moment.

• A rules file that will not parse stops everything

  A rules file that will not parse stops everything

  A silently empty ban list would pass every check.

• Signed, then re-read

  Signed, then re-read

  The callback verifies Slack’s signature and its freshness, and re-reads the draft’s status, because a Slack button stays clickable after it is pressed.

• Built, not yet deployed. Every file imports cleanly and the single path to publishing was verified by walking the graph; there has been no run against real Slack and Buffer accounts.
• Clip rendering is not built. The pipeline produces verified start and end times; rendering vertical clips needs FFmpeg on a host, or a paid render API — a decision with cost.
• Only the first draft of a set is published, through Buffer. Fanning out per platform is one node’s work, once the platforms are known.
• Transcription is AssemblyAI. The rejection-note loop is wired but not yet fed: the approval card still needs a field for the note.