Hosted Agent Infrastructure

What it does

This skill documents patterns and operational practices for running agents in hosted, sandboxed environments. It explains how to prebuild images, maintain warm pools, snapshot sessions, and extract outputs (branches/PRs/logs) so agents can perform reliable, low-latency background work. Concrete use cases include background coding agents that open PRs, collaborative sessions across clients, and self-spawned sub-agents for parallel tasks.

When to use it

Use this skill when designing or operating hosted agent runtimes: when you need predictable sandbox spin-up times, session snapshot/restore, per-session state isolation, user-attributed commits, or streaming feedback to clients (web, Slack, VS Code). Activate it instead of adjacent skills like harness-engineering or tool-design when the problem is the hosted runtime itself.

What's included

• Scripts: yes — the repo includes scripts and examples for image/warm-pool patterns (has_scripts=true)
• References: yes — architecture and integration references included (has_references=true)
• Instructions: step-by-step guidance on image build cadence, snapshot timing, warm-pool sizing, auth flows for user-based commits, and client integration patterns.

Compatible agents

Best suited for server-first agent frameworks and tools that can manage sandboxes and Git flows (e.g., Codex/Copilot-style automation, Cursor/Claude Code backends, or custom server agents that can call sandbox APIs).