agentleFS
tamtree-aitamtree-ai/stick-stage/skills/shorts-editor/SKILL.md

shorts-editor

Skill2 starsChanged 15 days agoMITView on GitHub

Cut a customer's long video into 9:16 shorts and act on review notes — pick the shorts from the transcript, write shorts.json, snap, frame, check the contact sheets, render drafts, and turn each review comment (with its time range) into an edit of shorts.json or framing overrides, then re-render and reply. Use when asked to make shorts from footage, a podcast, an interview, a film or a long video, or to apply review comments on such shorts. Never approves or releases a short.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run uv run shorts transcript --work $W and read all of it before choosing.

    line 14
  2. Very specific:

    After the edits: snap, frame, sheet (look), props, render the draft, and reply to each comment in one line saying what changed ("Starts on …

    line 75
  3. Distinctive:

    use footage outside the customer's source.

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Shorts editor

You are the editor's hands for the shorts service. The pipeline is packages/shorts (Python) plus the ShortFromFootage composition (src/app/shorts/). Read packages/shorts/README.md for the commands. Every step's output is a JSON file in one work dir, and you edit those files directly.

Picking shorts (writing shorts.json)

Run uv run shorts transcript --work $W and read all of it before choosing. A short is good when:

  1. It is a complete thought. It has a setup and a payoff. Never end before the explanation; that is the top complaint about AI clippers.
  2. It works cold. No "as I said earlier" or names the viewer hasn't met, unless the clip itself introduces them.
  3. It hooks in the first 2 s. Either the main segment opens strong, or add a teaser segment: 2–5 s of the payoff line, played first, then the main segment from the start.
  4. 20–60 s in total, teaser included. If the moment is longer, split it into several main segments that skip the slack. Each jump is a cut with an audio fade.
  5. Not a repeat. Shorts from the same video shouldn't share their payoff.

For each short write title (≤ 6 words, on screen after the teaser), hook (on-screen text over the opening, ≤ 8 words, a question or a claim), closing (the end card line: the moral, the fact or the CTA), and why (two plain sentences for the reviewer: what the viewer gets and why it works cold). Then run shorts snap and fix every ! it prints.

Framing

shorts frame → shorts sheet, then look at every contact sheet (sheet-<id>.jpg, Read the image). The yellow box is the 9:16 window, and an orange border means letterbox. Fix bad spans with overrides in shorts.json, keyed by shot id:

  • the window is on the wrong person: {"target": "<speaker>"}, or a fixed {"x": 0.62}
  • nobody fits, a wide, a plate, a screen recording: {"layout": "letterbox"}

Re-run frame and sheet after each change. Keep overrides: they survive a re-plan.

Rendering

shorts props --work $W --src-url <source path> then pnpm tsx scripts/short.ts $W/props-<id>.json --draft for review rounds (540×960, fast). Render final (no --draft) only after approval.

Review happens in t-asset

Drafts go to t-asset (~/sites/tasset, see its docs/shorts-review.md) as internal versions: only the studio sees them until the owner releases them from Drafts to review.

cd ~/sites/tasset
pnpm shorts ingest $W --videos <render dir> [--suffix -draft] --client "<client>" --project "<project>" --note "<what changed>"
pnpm shorts comments $W        # open notes on each short's newest version, JSON, times on the short's clock

pnpm fake-tamtree (or the real Tamtree flow) must be running so t-asset makes the previews.

Acting on review comments

Each comment has a time range in the short's timeline (at, until in the JSON). Map it to the source with framing.json's spans (out0, src0): source time = src0 + (t - out0).

The note saysEdit
start later / earlier, cut this bit, too longsegment t0/t1, or split a main in two around the cut; then snap
a different hook / openinga new teaser segment, or new hook text
wrong person in frame, cropped headoverrides[shot] target, x, or letterbox
caption word wrongfix the word's text in analysis.json words (keep the times)
title, closing line, CTAthe field in shorts.json
anything you can't do from these files (reshoot, new graphics, music, a logo we don't have)don't guess: reply that it needs the editor, and say what's needed

After the edits: snap, frame, sheet (look), props, render the draft, and reply to each comment in one line saying what changed ("Starts on her answer now, 0:03 shorter").

Never

  • approve, release or publish a short. People do that in t-asset.
  • change the words a speaker said, other than fixing a mis-heard caption.
  • use footage outside the customer's source.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.