hosted-agents

Pre-build environment images on a regular cadence (around 30 minutes) with dependencies installed and caches warmed, then maintain a pool of pre-warmed sandboxes. Start warming a sandbox predictively when the user begins typing, not when they submit the prompt.

Use GitHub App installation tokens only for cloning inside the sandbox, then set git config user.name and user.email to the prompting user. The sandbox pushes the branch and notifies the API, which creates the PR using the user's OAuth token.

Yes, reads are safe before sync completes because recently changed files rarely overlap with the agent's research targets. Block only writes until synchronization finishes to prevent conflicts, queueing pending writes in the meantime.

Modal Sandboxes provide near-instant startup and filesystem snapshots, while Cloudflare Durable Objects handle per-session state with WebSocket hibernation. The patterns in this Skill are provider-agnostic and adapt to Fly.io or similar platforms.

Work is lost when sandboxes are recycled or time out before results are extracted. Always snapshot the filesystem before termination and explicitly extract artifacts such as branches, PRs, and files during session teardown.

Avoid it when the task concerns the autonomous research loop, novelty gates, or merge boundaries, which belong to harness engineering, or when choosing agent topology without infrastructure concerns. This Skill covers only the hosted runtime layer.