sandcastle

Orchestrate AI coding agents in Docker sandboxes with parallel execution.

10|3|Updated Jan 23, 2026
One-click install
npx skills add https://github.com/av/skills --skill sandcastle
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: sandcastle
Source: https://github.com/av/skills/tree/main/sandcastle
Command: npx skills add https://github.com/av/skills --skill sandcastle

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

The sandcastle Skill unit addresses the challenges of coordinating AI coding agents across iterative tasks, particularly when dealing with isolated environments, parallel processing, and structured outputs.

Core Features & Use Cases

  • Sandbox Orchestration: Manage and run AI coding agents (like Claude Code, Codex, OpenCode) in Docker/Podman containers or in isolated worktrees.
  • Parallel Agent Management: Run multiple agents simultaneously in separate workspaces, allowing for efficient parallel processing.
  • Iterative Workflows: Create and manage iterative agent loops, including the ability to specify maximum iterations and branch strategies.
  • Structured Output Extraction: Capture structured data outputs from agents, allowing for better analysis and integration.
  • Interactive and Automated Sessions: Start interactive sessions for human-in-the-loop tasks or automate processes with scriptable run commands.

Quick Start

Use the sandcastle skill to initiate an iterative review cycle with Claude Code, using a Docker container sandbox. Run: npx skills run av/skills --skill sandcastle --interactive

Frequently Asked Questions about sandcastle

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run AI coding agents in isolated Docker sandboxes for parallel code refactoring?

You can orchestrate AI coding agents in Docker or Podman sandboxes to execute parallel code refactoring tasks. This skill manages isolated environments, enabling simultaneous agent execution and iterative development cycles.

What is the best way to capture structured output from multiple AI coding agents?

Capturing structured output from multiple AI coding agents is handled during parallel sandbox orchestration. The system automatically extracts structured data from each agent, allowing for better analysis and integration of the results.

Does this sandbox orchestration tool support interactive human-in-the-loop coding sessions?

Yes, sandbox orchestration supports interactive human-in-the-loop coding sessions. You can start interactive sessions using an `--interactive` flag, or automate processes entirely with scriptable run commands for autonomous agent workflows.

Can I manage iterative AI code generation workflows with maximum iteration limits?

Yes, you can manage iterative AI code generation workflows by creating agent loops with specified maximum iterations. This allows controlled, repetitive execution cycles suitable for bug hunting and code generation tasks.

Do I need Docker to use this AI agent orchestration skill, or are there other supported isolation methods?

Docker is not strictly required for AI agent orchestration. While it supports Docker and Podman containers for isolation, you can also run agents in isolated Git worktrees as an alternative environment setup.