← 返回 Skills 市场

Alfred OpenShell Sandbox

Name: Alfred OpenShell Sandbox
Author: lllljokerllll

作者 lJokerl · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

总下载

当前安装

版本数

在 OpenClaw 中安装

/install alfred-openshell-sandbox

功能描述

Provides isolated sandboxes using NVIDIA OpenShell for secure code execution, security scans, debugging, and test running with resource and network restricti...

使用说明 (SKILL.md)

OpenShell Sandbox Skill

Secure execution environment for specialist agents using NVIDIA OpenShell.

Overview

OpenShell provides sandboxed containers with Landlock LSM + seccomp + network namespaces + L7 policy engine. Each specialist agent gets an isolated sandbox for safe code execution.

Sandboxes Available

Sandbox	Agent	Purpose	Status
`coder-sandbox`	coder	Code execution, builds, tests	Ready
`security-sandbox`	security	Pentesting, security scans	Ready
`debug-sandbox`	debug	Bug reproduction, diagnosis	Ready
`test-sandbox`	qa-tester	Test execution	Ready

CLI Reference

# List all sandboxes
openshell sandbox list

# Execute command in sandbox
openshell sandbox exec -n \x3Csandbox-name> -- \x3Ccommand> [args...]

# Interactive shell
openshell sandbox connect -n \x3Csandbox-name>

# Create new sandbox
openshell sandbox create --name \x3Cname>

# Delete sandbox
openshell sandbox delete \x3Cname>

# View logs
openshell logs -n \x3Csandbox-name>

# Gateway status
openshell status

# Diagnose issues
openshell doctor check

Agent Integration

For Coder Agent

When executing code that could affect the host system:

# Instead of running locally:
python3 script.py

# Run in sandbox:
openshell sandbox exec -n coder-sandbox -- python3 /workspace/script.py

For Security Agent

When running security tools or scans:

# Run nmap, nikto, etc. in isolated sandbox
openshell sandbox exec -n security-sandbox -- nmap -sV target

For Debug Agent

When reproducing bugs or testing fixes:

openshell sandbox exec -n debug-sandbox -- node test.js

For QA-Tester

When running test suites:

openshell sandbox exec -n test-sandbox -- pytest tests/

File Transfer

To copy files between host and sandbox:

# Copy file INTO sandbox (via exec cat)
cat local_file.py | openshell sandbox exec -n coder-sandbox -- tee /workspace/local_file.py

# Copy file FROM sandbox
openshell sandbox exec -n coder-sandbox -- cat /workspace/result.txt > local_result.txt

Policies

Default policies apply L7 network restrictions. To view/modify:

openshell policy list

Resource Limits

CPU: Shared with host (24GB RAM server)
Network: Restricted by L7 policy (no outbound by default)
Disk: Ephemeral (deleted with sandbox)
Timeout: 30 min default per exec command

Troubleshooting

Sandbox not found: Run openshell sandbox list to check status
Gateway down: Run openshell status and openshell doctor check
Permission denied: Sandboxes run as unprivileged user
Network blocked: Default policy denies outbound; use openshell policy to modify

Architecture

Host (Ubuntu ARM64)
  └── OpenShell Gateway (Docker + k3s)
       ├── coder-sandbox (aarch64, Python 3.13, Node 22)
       ├── security-sandbox (aarch64)
       ├── debug-sandbox (aarch64)
       └── test-sandbox (aarch64)

Version

OpenShell CLI: 0.0.35
Base image: ghcr.io/nvidia/openshell-community/sandboxes/base:latest
Platform: aarch64 (ARM64)

安全使用建议

This is an instruction-only skill that documents how to use an existing NVIDIA OpenShell installation. Before installing or using it: (1) Verify you actually have the 'openshell' CLI and gateway reachable from the agent (the SKILL.md assumes this but the metadata does not declare it). (2) Confirm the OpenShell gateway and base image (ghcr.io/nvidia/openshell-community/...) are from a trusted source and up-to-date, and inspect their manifest/content if possible. (3) Validate sandbox isolation (Landlock/seccomp/L7 policies) in your environment — the skill will run arbitrary commands inside sandboxes and can copy host files into them, so a misconfigured gateway could leak host data. (4) Be cautious about running network scans or pentesting commands (legal/ethical considerations and possible outbound network requirements). (5) If you want this skill to be safer, request the author add explicit required-binaries and environment preconditions (openshell CLI presence, gateway URL/credentials if needed) and include provenance/homepage/source information.

功能分析

Type: OpenClaw Skill Name: alfred-openshell-sandbox Version: 1.0.0 The skill bundle provides documentation for 'OpenShell,' a sandboxing utility designed for secure code execution. The SKILL.md file contains standard CLI instructions for managing isolated environments (e.g., openshell sandbox exec) and lacks any indicators of malicious intent, data exfiltration, or prompt injection attacks.

能力评估

ℹ Purpose & Capability

The name/description (providing OpenShell sandboxes) matches the SKILL.md content: commands, sandboxes, policies and examples are consistent with that purpose. However, the skill metadata declares no required binaries or install steps even though every runtime example requires an 'openshell' CLI and an OpenShell gateway; that undeclared dependency is an incoherence.

✓ Instruction Scope

The SKILL.md confines actions to managing and running commands inside sandboxes (exec, connect, policy, logs). It does not instruct reading unrelated host files or environment variables. It does show file copy patterns (cat | exec tee) which legitimately transfer host data into sandboxes — this is expected for the stated purpose but means host files can be exposed to sandboxes if used.

✓ Install Mechanism

No install spec is present (instruction-only), which minimizes disk writes or hidden downloads. This is coherent for a skill that simply documents how to use an existing OpenShell installation — but the skill should have declared that the 'openshell' binary and gateway must be present and reachable.

ℹ Credentials

The skill requests no environment variables or credentials, which is proportional. However, it implicitly requires agent access to an OpenShell gateway/CLI and network access to targets for scans; those runtime privileges are not declared. Also it references a base image (ghcr.io/nvidia/openshell-community/...) — you should validate that image and the gateway's configuration before trusting sandbox isolation.

✓ Persistence & Privilege

The skill is not always-enabled and does not request persistent or system-wide configuration changes. It does not attempt to modify other skills' configs in the instructions.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install alfred-openshell-sandbox
安装完成后，直接呼叫该 Skill 的名称或使用 /alfred-openshell-sandbox 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Initial: 4 sandboxes (coder, security, debug, qa-tester), CLI reference, agent integration guide

元数据

Slug alfred-openshell-sandbox

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题