dsh-fetch-file

The built-in web_fetch is great at extracting article text for the model, but the moment a response is not a readable page it discards the raw byte stream entirely. Images, archives, PDFs, audio, installers are exactly the binary assets a model task often needs saved to disk, yet web_fetch simply throws them away. dsh-fetch-file closes that gap.

It takes an http/https URL, follows redirects, and streams the raw bytes straight to a file inside the workspace, returning the absolute path, actual byte count, HTTP status code, and Content-Type. Pure ESM, zero dependencies, no build step. The per-download cap defaults to 200 MB, the fetch timeout is 120 seconds, and a failed write cleans up its temporary file so no half-written artifact is left behind. A path fence guarantees the destination always resolves within the workspace root, and a self-check runs at mount time performing a real download of example.com and verifying that an out-of-bounds path is rejected, with evidence printed to the log.

It is aimed at developers building agent workflows, automation pipelines, or any system where the model must retrieve and operate on binary files directly. Use it when you need to grab a file, and stick with web_fetch when you need to read an article.