Communitygithub.com

loom-to-sop

Turn a Loom (or any screen recording) transcript into a written SOP someone else can follow the first time - the outcome, when to use it, tools and access, numbered one-action steps with the video timestamp for each, the quality check, common mistakes, and a list of gaps the video never explained. Use when someone pastes a Loom transcript, a .vtt or .srt caption file, or asks to "write an SOP" or "document this process".

What is loom-to-sop?

loom-to-sop is a Claude Code agent skill that turn a Loom (or any screen recording) transcript into a written SOP someone else can follow the first time - the outcome, when to use it, tools and access, numbered one-action steps with the video timestamp for each, the quality check, common mistakes, and a list of gaps the video never explained. Use when someone pastes a Loom transcript, a .vtt or .srt caption file, or asks to "write an SOP" or "document this process".

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/MairoMBaterzal/va-ai-kit/tree/HEAD/skills/loom-to-sop

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Loom to SOP

The client records a task once. You hand back an SOP the same day. The AI writes the first draft; you make it true by testing it.

What you need

  • The transcript. In Loom: open the video, click Transcript, copy it all (with timestamps if Loom shows them). A downloaded .vtt or .srt caption file works too.
  • The video link, if they have it (it goes at the top of the SOP).
  • Who will use the SOP (a new VA, the client's team), if known.

The framework: 8 parts, always in this order

  1. Title: verb + object. "Publish a blog post in WordPress", not "Blog SOP".
  2. Outcome: one sentence. What is true when this is done.
  3. When to use it: the trigger ("every Monday", "when a new client signs").
  4. Tools and access: every app, login and file the steps touch. Write "ask [client] for access to X" if the video uses something the reader may not have.
  5. Before you start: what must be ready first.
  6. Steps: numbered. One action per step, starting with a verb, written for someone who has never seen the screen. Add the exact button or menu name in bold, and the timestamp from the video in brackets, like [02:14], so the reader can watch that moment. Group steps under short headings when there are more than 10. Turn "and then you just..." into its own step.
  7. Check it's right: how to know the task worked (what to look at, what it should say).
  8. Common mistakes and what to do: every "be careful", "don't", "if this happens" from the video, as "If X, then Y."

Then, at the end, add a separate section the client sees:

Gaps: what the video didn't say. List every place where a step is unclear: a password not mentioned, a click done off-screen, "the usual settings", a decision made without saying why, a number with no source. Write each as a question for the client. This list is the most valuable part: it is the thing AI can't fill in for you.

How to write it

  • Use only what is in the transcript. Never invent a step, a setting, a link or a reason. If the speaker skips something, it goes in Gaps, not in the steps.
  • Remove filler ("um", "so basically", "let me just"), keep every action.
  • Short sentences. Second person ("Click Save"). No jargon the client didn't use.
  • Save it as output/sops/<title-in-kebab-case>.md. It pastes cleanly into Google Docs or Notion.

Check it (the VA's job)

Tell the person to do this before sending it: follow the SOP step by step with the video closed. Any step they had to guess becomes a fix or a Gaps question. Then send the SOP with the Gaps list on top of the message: "Here's the SOP. 3 questions before it's final."

Related Skills