What problem does it solve?
OpenScientist agents need isolated, reproducible execution environments to access tools and custom toolchains without polluting the host. This skill provides managed Docker sandboxes that can be concurrently active, bound to per-space mounts, with clear activation semantics and workdir transparency.
Core Features & Use Cases
- Per-space sandbox isolation: run toolchains inside containers bound to the agent's workspace, ensuring file path parity with the host.
- Dynamic activation: activate.sh --mount <path> or deep-run activations allow agents to attach or attach-on-demand to sandboxes.
- Workdir-aware execution: exec.sh enforces binding constraints and preserves host-dir semantics, enabling predictable behavior for host tools and agent workflows.
Quick Start
Activate a sandbox and run a command inside it for your agent workflow.