agentleFS
tamtree-aitamtree-ai/stick-stage/skills/topic-to-video/SKILL.md

topic-to-video

Skill2 starsChanged 15 days agoMITView on GitHub

Turn a topic into a finished ~30 second StickStage comedy video, end to end. Input is just a topic ("group chats", "returning a gift", "Monday meetings"); output is a postable MP4 in out/posts/<id>/ plus the script. Asks StickStage for the writer prompt, writes only the words, stages them with the skit-director skill, renders, self-checks and packages. Use when asked to make, generate or create a video/skit/short about a topic.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. addsumtech/topic-to-video★ 3same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository13 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 3 are distinctive, and 7 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 7

    Very specific

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

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

    Run everything from the repo root (the folder with package.json).

    line 15
  2. Very specific:

    Write public/skits/<id>/brief.json.

    line 26
  3. Very specific:

    Read system and prompt. Write one JSON object that answers that prompt and nothing else: no schema version, no roles, no cameras, no timing.

    line 40
  4. Very specific:

    Look at the contact sheet out/<id>-sheet.png (read it as an image) and use the skit-director checklist: faces readable, text not covering …

    line 80
  5. Distinctive:

    Keep the reply short:

    line 95
  6. Distinctive:

    Don't claim it's funny. A person judges that by watching it.

    line 104
  7. Distinctive:

    If the person asks for changes ("punchier ending", "make June lose it"), ask for a lines-only prompt and apply the reply.

    line 106
  8. Very specific:

    Write stdout over skit.json.

    line 114
  9. Very specific:

    Do not invent a series. If the person names one, read src/data/series/<id>.json and set meta.series (id, season, episode) on the skit.

    line 118
  10. Very specific:

    Read src/data/characters/<id>.md before writing those.

    line 128

≈ “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

Topic → 30 s video

The person gives you a topic. You deliver a ~30 s two-person skit as a postable MP4, without asking questions along the way. You own every step: premise, script, staging, render and packaging.

StickStage writes the prompt and turns the reply into a premise. You write only the words the prompt asks for. skit-director still never writes or changes a line. Everything after the premise follows skills/skit-director/SKILL.md. Read that file before staging.

Run everything from the repo root (the folder with package.json).

0. Before you start

  • The house pair is Milo and June. Lila and Theo (kids) and Moss and Dash (turtle and hare) are in the catalog when the bit needs them. The writer prompt already says how each one is played.
  • Pick a skit id: a lowercase slug of the topic, e.g. group-chat. If public/skits/<id>/ already exists, add -2, -3, and so on. Never overwrite another skit.

1. Ask StickStage for the words, then write only those

Write public/skits/<id>/brief.json. A brief is the topic and the cast, plus anything you have already decided. Leave template, set and scenes out when you want the reply to choose them.

{
  "topic": "group chats",
  "cast": [ { "id": "milo", "character": "milo" }, { "id": "june", "character": "june" } ]
}
pnpm write prompt --brief=public/skits/<id>/brief.json

Read system and prompt. Write one JSON object that answers that prompt and nothing else: no schema version, no roles, no cameras, no timing. Save it as public/skits/<id>/reply.json.

pnpm write draft --brief=public/skits/<id>/brief.json --reply=public/skits/<id>/reply.json

Stdout is the premise. Write it to public/skits/<id>/premise.json. Warnings on stderr (mood, line-count, set-fallback, …) are corrections StickStage already applied; keep going.

Exit 2 means the reply cannot be used. Stderr is one repair prompt. Answer that prompt once, overwrite reply.json, and run draft again. Do not invent a third try.

2. Stage it

pnpm new <id>              # premise.json → draft skit.json

Then refine public/skits/<id>/skit.json exactly as skills/skit-director/SKILL.md describes: comedic pauses before the turn and the punchline, one silent reaction beat at the key realization, one gesture per line anchored on its word, 1–3 SFX, and leaving shot out unless it's needed. From this step on, the line text is fixed. Staging never rewrites it.

3. Render and check (loop, at most 4 rounds)

pnpm direct <id> --say --json

Read public/skits/<id>/generated/direct.json:

  • Fix every validation or compile diagnostic at the path it names.
  • Fix every check finding at the error level, and deal with the warnings (see skit-director).
  • Length: compile.durationSec must land between 26 and 33 s.
    • Too short: first add pauses or a silent beat where the joke needs air. If it's still more than 3 s short, go back to step 1 and add an escalation line (edit premise.json, then run pnpm new <id> --force and redo the staging).
    • Too long: cut the weakest escalation line in premise.json the same way. Never trim the hook or the punchline.
  • Look at the contact sheet out/<id>-sheet.png (read it as an image) and use the skit-director checklist: faces readable, text not covering faces, the punchline has its camera event, the reaction face reads.

4. Package

pnpm batch <id>

This writes out/posts/<id>/<id>-<slug>-<hash>.{mp4,srt,txt,json}. The .txt file holds the post description, hashtags, the AI-voice note and the script.

5. Hand back

Keep the reply short:

  • The MP4 path (in out/posts/<id>/) and its duration.
  • The script, one line per row with the speaker, so the person can judge the joke without watching.
  • The angle in one sentence, and one or two staging choices worth knowing about.
  • Voices: this render uses the macOS say dev voices. For a version to post, the tamtree harness generates real TTS (voice.json, see docs/voice-contract.md) and runs pnpm render <id>. Then run pnpm batch <id> again.
  • Don't claim it's funny. A person judges that by watching it.

If the person asks for changes ("punchier ending", "make June lose it"), ask for a lines-only prompt and apply the reply. Staging you already did stays, except on lines whose words changed.

pnpm write revise --skit=public/skits/<id>/skit.json --note="punchier ending"
pnpm write apply --skit=public/skits/<id>/skit.json --reply=public/skits/<id>/reply.json

Write stdout over skit.json. Exit 2 is one repair pass, the same as draft. Then repeat steps 3–5.

Series

Do not invent a series. If the person names one, read src/data/series/<id>.json and set meta.series (id, season, episode) on the skit. The post manifest carries it. This skill still does not post. Shipped shows: milo-june, park-fables.

When they ask for kids, a fable, or three people

Milo and June stay the default. Do not add a seventh character.

  • Kids or a family bit: template family. Cast one of Milo or June, plus Lila, Theo, or both. Set classroom-1 or kitchen-1. The kid asks. The adult answers badly.
  • A fable: template fable. Cast Dash, then Moss. Set park-1. Dash is sure and leaves. Moss says the last true line.
  • Three people: template trio. Set living-1.

Read src/data/characters/<id>.md before writing those. An explainer (explainer) needs the cast to say the hook and the twist, and one narrator line for the concept. Keep narration under about 60% of the spoken time.

Failure modes

  • pnpm direct says a voice is missing or stale: you forgot --say.
  • Rhubarb or ffmpeg errors: report the exact error. Don't work around it by skipping the mouths.
  • You can't get within 26–33 s after 4 rounds: deliver the closest version and say how long it is.

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.