dsh-vidfetch

Before dsh-vidfetch, pulling a remote video into a DeepSeek Harness workspace meant stitching together shell scripts or calling yt-dlp by hand. This plugin collapses that into a single tool invocation: the agent passes a url and an output directory, and the plugin returns the best available audio-video stream already written to disk, or a concise error string when the URL cannot be resolved. No config files, no API keys, no extra setup on the agent side.

The surface area is deliberately small. One registered function, fetch_video, with two required parameters. Under the hood each call spawns a yt-dlp subprocess that inherits the host environment, so the only runtime requirement is Node 22.19 or later (or 24+). The plugin sends no telemetry and contacts no external service of its own; it simply proxies the model request into a local download.

It is well suited to developers whose Harness workflows need the agent to autonomously fetch video assets. Think content archiving, pre-processing for transcoding, or building a local media library. A practical caution: point dir at a sandboxed or scratch folder so the model does not decide write paths arbitrarily, and treat every downloaded file as untrusted content that should not flow directly into execution pipelines.