Skills

cinetic: Let Coding Agents Direct Cinematic Launch Short Films

cinetic is best when you want your coding agent to turn a one- or two-sentence brief directly into a cinematic launch short film or motion design — but only if your machine first has Node 22+, ffmpeg 6+, Python 3.11+, and headless Chromium in place, otherwise the rendering pipeline cannot run.

Official source · Leonxlnx/cinetic

Install

npx skills add Leonxlnx/cinetic

Who It's For

  • Anyone who wants to produce a short film for a product launch, brand promotion, or motion design project, but prefers working in a code environment and wants a coding agent to handle the whole process end to end.
  • Requires macOS or Linux, plus a machine that already has: Node 22+, ffmpeg 6+ (with libx264/ffprobe), Python 3.11+ (officially listed dependencies include numpy, scipy, soundfile, pyloudnorm, opencv-python, librosa, etc.), and a headless Chromium.
  • Not suitable when: on Windows without the runtime requirements above, official evidence does not state whether it works — treat that as unknown. If you only want pure text writing or document layout, this skill is irrelevant — its output is a rendered video master, not text. Brands with hard output-style requirements should note: the skill has built-in mandatory style bans against an AI-generated look (e.g., serif fonts, orange/amber, neon/glow, emoji), enforced by a linter and critics; your own brand guidelines take higher priority and can override these bans.

Your First Attempt in 3 Minutes

  1. Prerequisites: Confirm the runtime components listed above (Node 22+, ffmpeg 6+, Python 3.11+, headless Chromium) are installed; these are the official Requirements — missing any one will make rendering fail later.

  2. Install: In your project directory, run:

    npx skills add Leonxlnx/cinetic

    For Claude Code, the skill is installed into .claude/skills/cinetic (project-level); add -g to install into ~/.claude/skills/cinetic. The same command also supports other agents supported by the skills CLI (e.g., Codex, Cursor); use the -a flag to choose the target agent.

  3. Trigger the first task: Send the agent (with the skill installed) a brief of one or two sentences, e.g., describing the theme and tone of your product launch short film. For the first attempt, keep the duration and scope as small as possible, because the whole pipeline renders video, composes a soundtrack, and runs QA; whether it writes intermediate artifacts to your filesystem is not itemized in the official evidence and is unknown — run in a controlled directory.

  4. Signal of success: The agent builds the film project from your brief, by default using Remotion (React), or HyperFrames (HTML + GSAP); a single beat grid drives the visuals, the composed soundtrack, and QA, ultimately producing a rendered master that has passed the review gate. Seeing the rendered video master in your project is direct evidence the pipeline works.

Note: this is a video generation skill, with no separate text-review workflows for fiction or professional documents; if the agent's output deviates from expectations, keep iterating on the brief based on feedback from its built-in critics and linter.

Use Cases

Case 1: Product launch short film. Scenario: you're about to launch a new tool and need a short film shareable on social media. Send the agent a brief one-liner (product name, one-sentence value proposition, desired tone); per the official workflow, it builds the film project from the brief, constructs it with Remotion, drives visuals and the composed soundtrack from a unified beat grid, and delivers a rendered master. This much is confirmed: the duration cap, resolution, and other parameters are not stated in the evidence and must be observed in actual runs.

Case 2: Motion design with your own brand guidelines. Scenario: your team has strict visual standards and worries about an AI-generated look. Attach your brand requirements to the brief — the official stance is that "your brand always takes priority" and can override the built-in hard style bans. The agent builds under the linter and critics' checks, so the output won't fall into the default forbidden styles (serif fonts, orange/amber, neon glow, emoji, etc.). The expected result is a rendered short film matching your brand guidelines; the most effective file format for submitting brand guidelines is not stated in the evidence.

Boundaries of Use

  • Files and commands: Installation writes into the skills directory (.claude/skills/cinetic or the global path). During runs it executes the rendering pipeline (Remotion/HyperFrames builds, ffmpeg encoding, Python audio analysis) and necessarily invokes local Node, ffmpeg, Python, and Chromium; these are officially declared dependency behaviors. Beyond the install paths, which files it reads/writes and whether there are step-by-step confirmation points are not itemized in the official evidence — trial-run in a small scope first.
  • Network access: The evidence does not state whether the runtime needs network access (e.g., fetching dependencies or assets); it's unknown, so as a precaution test in a controlled environment first.
  • Environment prerequisites: Officially it targets macOS or Linux only; Node 22+, ffmpeg 6+, Python 3.11+, and headless Chromium are hard requirements.
  • Clients: Distributed via the skills CLI, supporting Claude Code (project-level and -g global), Codex, Cursor, and other agents supported by that CLI; platform support beyond these is not stated — don't assume it can be triggered in unlisted agents.
  • Unconfirmed key information: The output video's duration range, resolution and format, cost and rendering time, and failure fallback behavior are not given in the official evidence; rely on actual runs before use.