Git

A clone has three resource profiles — network, CPU, and writes. No single isolate budget fits all three, so each phase gets its own.

git clone fetches real packfiles from real remotes over HTTPS. It is one of the heaviest operations a session runs. Fetch needs network time, index-pack needs CPU, and checkout makes many writes. Each phase gets a fresh CPU budget, and checkout is split into resumable chunks.

The terminal below clones a real GitHub repository over HTTPS: packfile negotiation, indexing, and checkout, ending in ordinary files on the session disk.

nimbus — git

Live sandbox — connecting…

git clone https://github.com/AshishKumar4/Markflow
cd Markflow && npm install && npm run dev

facebook/react, with 7,300 files, clones in ~28 s. Bounded continuation has materialized 84,000-file worktrees. The result is ordinary session disk: the editor sees it, npm install runs in it, and it persists.

Transport is HTTPS only; SSH is not supported. There is also a known ceiling: index-pack CPU can still exceed one invocation’s budget on clones with 75,000+ objects, such as microsoft/TypeScript. Making that phase resumable is in progress.