What problem does it solve? Running AI coding agents directly on your host machine risks exposing neighboring projects, credentials, and system state. This Skill creates isolated, per-project Docker sandboxes so each agent works only within its intended scope. ## Core Features & Use Cases - Stack Inspection and Proposal: Inspects the project stack and proposes a sandbox with explicit packages, bind mounts, and a confirmation hash before anything is created. - Isolated Per-Project Containers: Creates one Linux sandbox per project and agent combination, supporting claude-code, codex, opencode, pi, and kimi-code, without exposing neighboring projects. - Lifecycle Operations: Provides start, stop, freshness-check, and clean commands for interactive TUI sessions inside the sandbox. - Use Case: You want Claude Code to work on your backend repo without access to your other projects. Ask for a coding sandbox, review the proposed binds and hash, confirm, and open an interactive session in the isolated container. ## Quick Start Ask the agent to set up a coding sandbox for this project and confirm the proposed configuration hash before it builds the container.
