srt

Run sandboxed Claude sessions with per-project .srt.json configuration.

11|5|Updated Jan 9, 2026

Search Tags:#security #Claude #CLI #runtime #sandbox #autonomous

Ready to run

npx skills add https://github.com/rbergman/dark-matter-marketplace --skill srt

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: srt
Source: https://github.com/rbergman/dark-matter-marketplace/tree/main/plugins/workflow/skills/srt
Command: npx skills add https://github.com/rbergman/dark-matter-marketplace --skill srt

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill provides a safe, isolated environment to run Claude-driven workflows, preventing unintended system changes and protecting sensitive data during CLI and autonomous tasks.

Core Features & Use Cases

Sandbox Runtime (srt) supports OS-level sandboxing for Claude prompts on macOS and Linux.
Per-project .srt.json configuration for predictable behavior and repeatable experiments.
Fine-grained network and filesystem controls to limit exposure and risk.
Use cases include DX testing, CI/CD autonomous runs, and isolated subagent tasks to improve reliability and safety.

Quick Start

Install the Sandbox Runtime tool globally: npm install -g @anthropic-ai/sandbox-runtime Run a sandboxed Claude session: srt -s .srt.json -c 'claude --dangerously-skip-permissions --no-session-persistence -p "prompt"'

Frequently Asked Questions about srt

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

FAQPage Schema

How do I sandbox Claude CLI autonomous workflows to prevent unintended system changes?▼

Sandbox Claude CLI autonomous workflows by applying OS-level isolation with the srt runtime and a per-project .srt.json configuration file to restrict network and filesystem access. This prevents unintended system modifications and protects sensitive data during execution.

What is the best way to run isolated prompts for developer experience testing and CI/CD pipelines?▼

The best way to run isolated prompts for DX testing and CI/CD pipelines is to execute Claude within a sandbox runtime. This provides fine-grained filesystem and network controls to limit exposure and ensure reliable, repeatable autonomous task execution.

Does the srt sandbox runtime support both macOS and Linux environments?▼

Yes, the srt sandbox runtime supports OS-level sandboxing for Claude prompts on both macOS and Linux environments. It enforces permission management through allow and deny controls to limit exposure and risk.

How do I configure allow and deny controls for an isolated Claude subagent task?▼

Configure allow and deny controls for an isolated Claude subagent task by defining fine-grained network and filesystem permissions in a per-project .srt.json file. This configuration ensures predictable behavior and repeatable experiments.

Do I need a specific tool installed to execute sandboxed Claude sessions with permission management?▼

Yes, you need to install the sandbox runtime tool globally via npm to execute sandboxed Claude sessions. You then pass the .srt.json configuration file to the srt tool to apply governance and isolation controls.

Why use a sandbox runtime for governance-heavy prompts instead of running Claude directly?▼

Use a sandbox runtime for governance-heavy prompts to enforce strict isolation and permission management. Running Claude directly risks unintended system changes, while the sandbox applies allow and deny controls to protect sensitive data during autonomous workflows.