swarmbox 0.1.0

Python orchestration for coding agents in local, container, and cloud sandboxes.

SwarmBox

SwarmBox is a Python port and expansion of Sandcastle-style coding-agent orchestration. It runs coding agents in git worktrees, Docker/Podman containers, optional cloud sandboxes, or directly on the host for interactive sessions.

The project is designed for Sandcastle feature parity first:

• run, interactive, create_sandbox, and create_worktree
• branch strategies: head, merge-to-head, and branch
• prompt files, {{ARG}} substitution, and shell expression expansion
• host and sandbox lifecycle hooks
• Docker, Podman, Vercel, Daytona, and no-sandbox providers
• Claude session capture/resume-compatible storage
• init templates for loop, review, and parallel planner workflows

SwarmBox also adds an agent registry seeded from SwarmVault and SwarmClaw so any coding-agent CLI can be detected or configured with a generic command adapter.

Quick Start

python -m pip install -e .
swarmbox agents list
swarmbox init --agent codex --sandbox docker --template blank --no-build

from swarmbox import codex, docker, run

result = run(
    agent=codex("gpt-5.4-codex"),
    sandbox=docker(),
    prompt_file=".swarmbox/prompt.md",
)
print(result.branch, result.commits)

Reference

The behavior and API shape are based on the upstream MIT-licensed Sandcastle project: https://github.com/mattpocock/sandcastle