@chenjie1129/dsh-remotion-video-plugin
用于 DeepSeek Harness 的 Remotion 视频创建和经验证渲染插件
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:chenjie1129/remotion-video-plugin#bc23671b7e8da036196925dbfcca69090c4403f2
Remotion Video Plugin 以 @chenjie1129/dsh-remotion-video-plugin 发布，当前版本为 0.4.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Give DeepSeek Harness a structured Remotion workflow and five workspace-confined tools for diagnosing projects, discovering compositions, rendering stills and videos, and verifying output metadata.
12-second Remotion Video Plugin demo
Watch the 12-second MP4 · Reproduce the proof
After 0.4.0 is published to npm, install the immutable version into the DeepSeek Harness Web profile:
dsh plugin --profile web add @chenjie1129/dsh-remotion-video-plugin@0.4.0
dsh web
Until npm publication is verified, use the pinned GitHub release tag:
dsh plugin --profile web add github:chenjie1129/remotion-video-plugin#v0.4.0
dsh web
Start a standard-agent session and ask:
Create a 15-second 16:9 product-launch video. Diagnose the project, use frame-driven animation, render a representative still, then render and verify the MP4.
remotion-video skill with eight focused rule modules.| Tool | Purpose | Writes files |
|---|---|---|
remotion_doctor | Diagnose package, entry point, CLI, ffprobe, and browser readiness | No |
remotion_list_compositions | Return registered composition IDs | No |
remotion_render_still | Render a PNG or JPEG frame | Yes |
remotion_render_video | Render and metadata-probe a video | Yes |
remotion_probe_output | Return normalized media metadata | No |
The skill and executable tools are separate Cordis rows, so an operator can disable either capability without editing the other.
The repository includes three distinct evidence levels:
| Evidence | What it proves | Current record |
|---|---|---|
| Unit and contract checks | Skill lifecycle, bundle rows, routed resources, eval cases, tool schemas, argv safety, path rejection, artifact presentation | Test source |
| Real Harness probe | Tool registry mount, managed subprocess execution, project doctor, bundled ffprobe, and checked-in MP4 metadata | Smoke results |
| Browser-backed integration | Composition discovery plus a fresh still and video rendered through all five tools | Integration workflow |
The checked-in demo is a real H.264 render at 1280×720, 30 fps, and about 12 seconds. A release is not declared fully proven until the browser-backed integration workflow passes for the release commit. Probe-only success is deliberately not presented as a render-tool pass.
^22.19.0 or >=24.0.0.0.2.0: dsh-skill, dsh-tools, and dsh-subprocess.Rendering executes the selected project's JavaScript and installed dependencies. Tool paths are confined to the active session workspace, commands use fixed argument arrays, output replacement is opt-in, process output and props are bounded, and cancellation reaches the managed process tree. The plugin itself collects no telemetry, stores no credentials, and uploads no media; project code, remote assets, or Remotion browser acquisition may still use the network. See Security and Tool contract.
Inspect the composed profile:
dsh --profile web --dump-config
It should contain both remotion-video-plugin and remotion-video-tools. A trusted browser can be configured on the tools row:
- id: remotion-video-tools
name: '@chenjie1129/dsh-remotion-video-plugin/tools'
config:
browserExecutable: /absolute/path/to/chrome-or-chromium
browserMode: chrome-for-testing
Remove both rows by uninstalling the bundle, then restart Harness:
dsh plugin --profile web remove @chenjie1129/dsh-remotion-video-plugin
npm ci
npm run check
npm run release:verify
npm run test:e2e -- --probe-only
Rendering changes must also pass the full npm run test:e2e command with a real Harness checkout and approved browser. See Compatibility, Smoke test, Release procedure, Contributing, and Changelog.
This is an independent community plugin, not an official DeepSeek or Remotion project. The plugin is MIT licensed. Remotion has separate license terms; users are responsible for checking them for their team and usage model.