v1.3.2: review tools, fewer rules

The style guides now describe each look and leave the story, structure, and extras to the agent. Two review tools help it check its own work.

Review tools

• Range stills: node render.mjs stills 12-14 renders six evenly spaced frames across a range, so motion a single still misses (something leaving the frame, a card that never appears) shows up in the contact sheet.
• Peak-aligned effects: each sound effect in audio.json now peaks at its cue time instead of starting there, so a whoosh or pop lands on the cut. A cue near 0 s is trimmed instead of dropped.

Style guides

• Dropped the bans on logos, characters, slogans, and brand names, the banned-effects list, the "only when / only after" gates, the plan-reporting steps, and the scene-by-scene structure tables copied from the originals.
• Kept the Signature tables, number sourcing, credits, and technical constraints.
• Free style: the agent works out the look however it sees fit and writes a short Signature to check its stills against.

Assets

No new videos; the README hero and per-style examples are the same as in v1.3.1.

v1.3.1: style fixes

We made a Claude Opus 5.5 launch video in every style, each by a fresh agent given only this skill and the topic, and fixed what broke.

Fixes

• Hero comic: a panel with halftone shading inside no longer inks every dot and turns the page black.
• 16-bit arcade: the HIGH SCORE value counts up to its real value under its label, instead of flashing random digits.
• CRT terminal: the banner font has %, $, and ,.
• Thermal receipt: the stamp sits beside the slip instead of over its values.
• Hand-drawn: the guide keeps the look and no longer lists the original's scenes; the story comes from the research.

Assets

CodeVideo_opus55.mp4 is the README hero: a hand-drawn Claude Opus 5.5 launch video made by a fresh agent with only this skill, in the style of @nahiddotai's original. The per-style examples (CodeVideo_<style>.mp4) and intro videos are the same as in v1.3.0.

v1.3.0: design, research, format

Give code-video three things and get an MP4: a design, the topic, and the format (length, frame, language). The agent does the research and writes the story.

What's new

• Designs instead of story formats. Pick a style, a brand preset, any site's DESIGN.md, or describe any look in your own words (or leave it to the agent). The story comes from the research.
• A lighter skill. SKILL.md is down from about 1,950 to 665 words with four hard rules. In our A/B test on three styles, a video cost 23–51% less with no visible drop in quality.
• New styles: hero comic, toon, scrapbook, torn-paper collage, particles, split-flap board, neon sign, and UI morph (after @twoclipping's UI motion study).
• Brand design language: Apple-, Samsung-, Ferrari-, Nike-, and Spotify-style presets, plus any site's DESIGN.md (for example from VoltAgent/awesome-design-md). Not affiliated with these companies; no logos, slogans, or proprietary fonts.
• Free style: describe a look, show a reference image or video, or let the agent invent one; it designs a signature first and shows it in the plan.
• Gallery: the README shows a clip from an example in every style.

Assets

One example video per style, each made by a fresh agent that had only this skill, about the skill itself: CodeVideo_<style>.mp4. CodeVideo_free.mp4 is a look the agent invented (a pen plotter). The intro videos are CodeVideo_intro_en.mp4 and CodeVideo_intro_ko.mp4.

v1.2.0 — reproduces its credited originals

The credited looks and formats now reproduce the videos they come from. Each one was tested on its original's own topic by a fresh agent that had only this skill; the README shows every original next to that reproduction.

What's new

• Reproduces the originals. New style modules draw each original's signature pieces: handdrawn.js (@nahiddotai), motion.js (@digitalstrategyai), sand.js (@Michaelzsguo), lyric.js (@goodside), beat.js (@twoclipping).
• Signature checklists. Every credited guide lists the traits of the original and the helper that draws each one. The skill looks at the original's frames before planning and lays its render next to them in QA (scripts/reference_sheet.py).
• Faster. Frames are captured as JPEG (2–3× faster than PNG, visually identical after H.264), and the QA render is the delivery file. Same-topic test runs took 4–13 minutes each, down from 15–23.
• Headless-safe. The skill waits for its own research and renders instead of ending its turn, so non-interactive runs finish.
• New kit pieces. Phone and cursor props, subframe motion blur (VIDEO.blur), QUERY=lang=ko for two-language renders, and all declared fonts loaded before the first frame.
• README. Original-vs-reproduction pairs, a Contributing section (new looks welcome), and no fixed look counts.

Assets

• CodeVideo_intro_en.mp4, CodeVideo_intro_ko.mp4: a 45-second intro made with this skill in the default hand-drawn look.

No original prompt text is included; the guides are our own write-ups, credited to their creators.

Install

npx skills add Changroro/code-video -g

v1.0.0

First release of the handdrawn-promo-video skill.

• Asks for length, aspect ratio, characters, and on-screen language first.
• Researches the topic with sources, gets the storyboard approved, then builds, checks, and renders an MP4.

Asset: SkillIntro.mp4, this skill's intro made with the skill (45 s, 1920×1080, no sound).

Credit: the hand-drawn style comes from @nahiddotai's "Introducing Opus 5.5" launch video and the prompt they shared.