Create Post: Video

Video is the highest-reach format on every platform Marky publishes to, and it is the one format agents historically couldn't produce. This skill closes that gap: HyperFrames renders the video, Marky brands and ships it.

Read the marky-api skill first for auth, endpoints, and the brand cache. Base URL https://api.mymarky.ai/api, header Authorization: Bearer mk_live_YOUR_KEY.

Check for HyperFrames

This skill depends on HyperFrames — an open-source system that renders video from HTML, built for agents.

• If the hyperframes skill is available in your session, use it: read /hyperframes first (it routes to the right workflow — motion graphic, explainer, product promo, captions) and follow its instructions to author and render.

• If it is not installed, tell the user how to add it, then continue once it is:

  npx skills add heygen-com/hyperframes
  

Do not try to render video any other way — no ffmpeg improvisation, no other frameworks. HyperFrames is the supported path.

Stage 1 — Brand the composition

Before authoring any frames, load the brand so the video looks like the business, not a template:

1. Read the brand profile — the injected brand-cache.md snapshot if fresh, else one get_business / GET /businesses/{id} call (which refreshes the cache).
2. Apply it to the HyperFrames composition:
   • palettes → the video's color system (backgrounds, accents, text).
   • header_font / body_font → typography.
   • logo_url → end-card or corner mark (respect logo_background_color).
   • tone → the on-screen copy's voice; ctas → the closing call-to-action.
3. Keep platform constraints in mind: 9:16 for Reels/TikTok/Stories, 1:1 or 16:9 for feeds; short (under ~30s) unless the user asks for longer.

Where the project lives. Unless the user names a location, create the HyperFrames project at ~/.marky/videos/<business_id>/<project-slug>/ — NOT under the current working directory. Video workspaces are big (tens of MB) and cwd could be inside someone's repo; ~/.marky keeps them out of git, out of iCloud-synced folders, and groups them per business so past projects become reusable templates (the brand-remixed frame.md and storyboards accumulate there). After rendering, delete the project's capture/ directory — it's regenerable scratch and usually the bulk of the size.

Stage 2 — Author and render

Follow the HyperFrames workflow for the video type the user wants (motion graphic, stat count-up, product promo, explainer, captioned clip). Show the user the plan — scenes, copy, duration — before rendering long pieces. Render to MP4.

Sound effects and music — free paths first

A video with no sound design feels unfinished. You do not need a paid account to fix that:

1. Bundled SFX library (free, offline, use this first). HyperFrames' media-use skill ships ~19 ready-to-use effects (whoosh, key-press, typing, click, chime, pop, riser, impact-bass, sparkle, glitch) at <skills-root>/media-use/audio/assets/sfx/. Resolve one by intent with node <media-use>/scripts/resolve.mjs --type sfx --intent "whoosh", or copy the file straight into your project's assets/sfx/ and cue it from the workflow's audio metadata. Keep SFX volume at 0.2–0.3 so narration stays on top.
2. HeyGen catalog (free-usage allowance). With npx hyperframes auth login (OAuth), SFX and background-music retrieval ride HeyGen's free allowance — signing in does not require a paid plan. The workflow's fetch-sfx / BGM steps then work out of the box.
3. CC0 web libraries (when the bundle lacks the sound). freesound.org (filter license to CC0) and pixabay.com/sound-effects are safe for commercial posts with no attribution. Download the file into assets/sfx/ and cue it like a bundled one.

Match effects to on-screen beats (a whoosh on the hook, key clicks under typing, a chime on the CTA) rather than scattering them; background music, if any, sits well under the voiceover (~0.2 volume).

Stage 3 — Upload to Marky

curl -X POST "https://api.mymarky.ai/api/businesses/BIZ_ID/media" \
  -H "Authorization: Bearer mk_live_YOUR_KEY" \
  -F "file=@/path/to/video.mp4"

Keep the returned original_url. Uploads cap at 50 MB — if the render is bigger, reduce duration or resolution and re-render rather than compressing into artifacts.

Stage 4 — Caption, approve, schedule

1. Write the caption in the brand voice (tone, caption_writing_rules, caption_suffix — see "Write like the business" in the marky-api skill).
2. Create the post with the caption and the uploaded media.
3. Show the user the video and caption, and get explicit approval. For multiple cuts or variants, use the review board in pick mode (${CLAUDE_PLUGIN_ROOT}/scripts/review-board.py items.json --mode pick with each variant's local file path or URL as media_url; run in background, parse BOARD_URL:, AskUserQuestion with the URL, read feedback.json). Append the result to ~/.marky/feedback-log.jsonl with context: "video-variants" and read it before styling the next video — see "The feedback log" in the marky-api skill. Never schedule or publish a video without it.
4. On approval, schedule via the schedule-posts skill. Video is supported everywhere — target every connected platform whose integration status is valid.
5. After the scheduled time, confirm publish_results shows success per platform.

Tag the post

Set metadata on the post when you create it — {"media_type": "animated-video" (or "talking-head" for captured footage), "created_by": "create-post-video", "hook": ..., "topic": ...} — per "Tag every post" in the marky-api skill. This is what lets future performance reviews compare this post's engagement against other media types.

Notes

• Rendering can take a few minutes for longer pieces — tell the user before starting.
• A video the user supplies (rather than one you render) skips straight to Stage 3. To COLLECT user footage (a talking-head clip with a teleprompter, a screen demo), use the capture studio — ${CLAUDE_PLUGIN_ROOT}/scripts/capture-studio.py (see Stage 4 of plan-social-content for the flow) — then hand the file to HyperFrames for captions / packaging, or straight to Stage 3.
• If HyperFrames fails to render, report the actual error and stop — don't silently fall back to an image post the user didn't ask for.