agentleFS
AnayDhawanAnayDhawan/openvidstudio/.github/skills/openvidstudio-director/SKILL.md

openvidstudio-director

Skill33 starsChanged 14 days agoApache-2.0View on GitHub

Use when making a peak or mid creative-level launch video, promo, trailer or "make it look great" demo with openvidstudio, where the video should be directed (a story, a designed look, an original score, hand-built scenes) rather than assembled from templates. Covers the whole process from intake to the final description file. Use the plain openvidstudio skill for literal, exactly-as-specified videos.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every number and claim in the video comes from a real capture or a real tool result.

    line 12
  2. Distinctive:

    Ask for the creative level (peak: full creative liberty, bespoke everything.

    line 17
  3. Distinctive:

    Use elicitation if the client has it, otherwise ask.

    line 19
  4. Distinctive:

    Ask only what the repo cannot tell you: length, the two or three things to show, the audience.

    line 28
  5. Very specific:

    Write narration to each beat's word budget.

    line 49
  6. Very specific:

    Run the product's own commands and tools for real outputs, then collect_facts to merge every tool result into src/videos/<name>/facts.json.

    line 57
  7. Very specific:

    Write src/videos/<name>/design.ts with defineDesignSystem, from the brief.

    line 63
  8. Distinctive:

    Look at the contact grid. Read the numeric findings that came with it.

    line 83
  9. Very specific:

    No em dashes, on screen or in narration.

    line 109
  10. Distinctive:

    Never invent a feature, number or quote.

    line 112

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

openvidstudio-director

The process for a directed video. Procedural on purpose: each step says what to do, which tool does the mechanical part, and what "done" means. The craft behind the steps is in docs/CREATIVE.md; read it before step 4 and again before step 9.

Every number and claim in the video comes from a real capture or a real tool result. If the product cannot show something, build a labelled sample, film it for real, and disclose it.

1. Intake: level and budget

  • Ask for the creative level (peak: full creative liberty, bespoke everything; mid: strong defaults, mostly authored scenes; literal: exactly what was specified) and a time budget. Use elicitation if the client has it, otherwise ask.
  • plan_production with both and the target length. Show the plan and its time estimate at the approval gate. If fitsBudget is false, show cutFirst and let the user choose.
  • Done when: level, budget and a measured time estimate are agreed.

2. Read the repo before asking anything

  • README, changelog, recent commit subjects, issue titles, the product's real outputs, the dev command. Start the app and confirm it responds.
  • Ask only what the repo cannot tell you: length, the two or three things to show, the audience.
  • preflight once. Done when it passes.

3. Brand

  • extract_brand with sourceRepo set to the repo. Fill in anything it lists as unresolved.
  • Done when src/brand.ts exists and its colours show in a rendered frame (step 9 checks this).

4. The brief

  • Find the story (CREATIVE.md §1). Write the logline, audience, the one feeling, the story device and why it fits, the arc (hook, tension, turn, proof, payoff, call to action), the design rationale (palette roles, motif and where it appears, type pair, motion personality), the music plan (bpm, key, section map, impact points), and a shot list marking each shot real or constructed with its focal point.
  • write_brief with the prose and the JSON. It validates shape and refuses em dashes.
  • Done when the brief is written. Show the user the whole brief and get an explicit yes first.

5. Beats and the approval gate

  • plan_beats, with snapToBars and the score's bpm when the video is cut to music.
  • Write narration to each beat's word budget. validate_beats.
  • Show the full beats file and wait for an explicit yes. Then write_beats_file.
  • Done when beats.json is on disk.

6. Captures and facts

  • A capture tool per beat: capture_screenshot, capture_screen_recording, capture_terminal, capture_desktop or capture_mobile. Add a wait interaction for entrance animations.
  • Run the product's own commands and tools for real outputs, then collect_facts to merge every tool result into src/videos/<name>/facts.json. Scenes import facts instead of retyping.
  • Done when every beat that needs a capture has one and facts.json exists.

7. Design system

  • Write src/videos/<name>/design.ts with defineDesignSystem, from the brief. Components and validators read it. Brand tokens seed it; they do not dictate it.
  • Done when scenes can import it and nothing needs a hard-coded colour.

8. Score

  • compose_score with mode: "bed" for narrated video or "music" for music-only. It writes the track, a cues file exact by construction, and a spectrogram, and reports band balance, clicks and per-section loudness in numbers. Read them; you cannot listen.
  • Or use a bed through find_music_bed and import_music_bed, then plan_music_cues.
  • Done when the checks are clean and cues exist.

9. Scenes and the review loop

  • Author scenes in scenes/ from @openvidstudio/core's primitives and design.ts, using useMusicClock(beatId) to land hits on the score and DirectedShot for captures. Start from scaffold_scene with mode: "starter" or "blank". Write your own component when the motif needs one. Mid level may use a template for a beat that does not need more.
  • Loop, until the rubric in CREATIVE.md §9 passes:
    1. render_stills with perBeat: "review" and the next round.
    2. Look at the contact grid. Read the numeric findings that came with it.
    3. Fix the worst problem. Repeat.
  • Done when a round has no errors and each rubric question has a yes.

10. Gates

  • validate_scenes with render: true: cropping, text size, reading time, contrast, em dashes, safe area, HUD overlap, dead frames, palette. Errors block the render. Warnings need a decision.
  • validate_beats again if the beats changed. generate_narration if the video has a voice.
  • stitch_composition. Done when every gate is green or consciously waived.

11. Render and assemble

  • render_video with draft: true, then full quality. It runs async; poll job_status.
  • finalize_video: loudness to -14 LUFS with a -1.5 dBTP ceiling, a poster frame, a size-capped share copy, and output/<name>-DESCRIPTION.md.
  • qc_extract_frames, and work docs/PIPELINE.md's checklist.

12. Hand over

  • Show the user the mp4 path, its length, the poster, and a contact grid.
  • Point at the description file. Say which shots are constructed, and what in it needs a human decision (a credit, a link). A render nobody looked at is not a delivered video.

Rules that never relax

  • No em dashes, on screen or in narration. No linear easing. Every scene inside CinematicScene. Motion never stops.
  • Nothing is written to disk or rendered before the user approves the brief and the beats.
  • Never invent a feature, number or quote. A missing shot beats a made-up one.

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.