A sandboxed automation tool that delegates development tasks to AI coding agents running in isolated containers.
tsk is a command-line tool that delegates software development tasks to AI coding agents running in secure, isolated container sandboxes. It solves the problem of safely integrating AI assistants into development workflows by preventing accidental damage to repositories, leaking API keys, or making unauthorized network calls. tsk automatically copies your repository, builds container images with your toolchain, and lets agents work asynchronously before fetching their changes back as git branches.
Developers and engineering teams using AI coding assistants like Claude Code or Codex who want to automate tasks safely without risking their codebase or data. It's particularly useful for those managing multiple parallel tasks, complex project setups, or requiring strict isolation for agent experiments.
Developers choose tsk because it provides robust safety guarantees through container isolation while automating the tedious setup of agent environments. Its unique parallel execution, task chaining, and template system enable scalable automation workflows that respect developer workflow and review cycles.
Keeping your agents out of trouble with sandboxed coding agent automation
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
tsk runs AI agents in filesystem and network-isolated containers with configurable domain allowlists, preventing accidental damage or data leaks as highlighted in the README's security features.
It auto-detects your project stack (Go, Rust, Python, etc.) and builds appropriate container images, reducing manual configuration effort for most projects.
The server daemon allows queuing multiple tasks for parallel execution with a TUI dashboard, enabling efficient handling of multiple AI agents simultaneously.
Supports chaining tasks so child tasks start from parent's commits, facilitating complex, multi-step development processes as described in the task chaining section.
Currently only supports Claude Code and Codex, missing integration with popular alternatives like GitHub Copilot or open-source LLMs, which restricts flexibility.
Requires understanding of Docker layers, TOML configuration files, and custom setup for non-standard stacks, with the README admitting it's 'still a work in progress' for ease of use.
Building and running containers for each task introduces latency and resource consumption, which may be prohibitive for quick iterations or resource-constrained environments.