Sandcastle open-sources sandboxed coding agents

This feels less like a flashy demo and more like missing infrastructure for serious agentic coding: if agents are going to touch real repos, sandboxing and branch-based merge-back need to be first-class.

• –The branch strategy is the real product here, not just the agent wrapper; it gives each run a contained path back into the main codebase.
• –Provider-agnostic support matters because teams can start local with Docker or Podman and later move to Vercel without rewriting the orchestration layer.
• –The repo reads like a practical toolkit for AFK agents, code review automation, and multi-agent parallelism, which is where this category starts to become useful.
• –Open-sourcing it lowers the barrier for teams that want agent workflows but don’t want to build sandbox plumbing, merge choreography, and setup scaffolding from scratch.