Show HN: TSK – agent sandbox, delegation, and parallelization tool (github.com)

TSK is a Rust CLI that lets you delegate development tasks to AI coding agents running in isolated Docker sandboxes and returns completed work as git branches for human review. It supports Claude Code and Codex agents, runs agents in parallel containerized environments (with a Squid forward proxy to restrict network access), and offers an orchestrator daemon to manage worker pools. Typical flows include tsk shell (interactive sandbox that copies your repo, creates a branch, and saves work back to the original repo) and tsk run or tsk add to launch autonomous agents with task-type templates, descriptions, and flags like --agent, --type, --name. Templates use placeholders (e.g., {{description}}) to reduce prompt boilerplate; outputs and logs live in ~/.local/share/tsk tasks directories. For AI/ML engineering this is significant because it operationalizes a “lead engineer + AI team” pattern: multiple LLM agents can work independently on the same repo, produce branches/pull-requests, and be evaluated using your normal Git/CICD workflow—improving parallelism, reproducibility, and auditability versus ad-hoc prompts. Key technical points: built in Rust, uses Docker for reproducible stacks (base OS, language stack, project layer, agent install snippets), configurable templates and proxy, and worker-managed parallelism. Caveats: agent support is limited today, Docker image customization can be required for complex projects, and sandboxing/credential handling remain important security considerations.