AI Skill Hub 推荐使用:沙箱探针 是一款优质的Agent工作流。AI 综合评分 7.2 分,在同类工具中表现稳健。如果你正在寻找可靠的Agent工作流解决方案,这是一个值得深入了解的选择。
沙箱探针 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
沙箱探针 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:go install(推荐) go install github.com/controlplaneio/sandbox-probe@latest # 方式二:从源码编译 git clone https://github.com/controlplaneio/sandbox-probe cd sandbox-probe go build -o sandbox-probe . # 方式三:下载预编译二进制 # 访问 Releases 页面下载对应平台二进制文件 # https://github.com/controlplaneio/sandbox-probe/releases
# 查看帮助 sandbox-probe --help # 基本运行 sandbox-probe [options] <input> # 详细使用说明请查阅文档 # https://github.com/controlplaneio/sandbox-probe
# sandbox-probe 配置说明 # 查看配置选项 sandbox-probe --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量(覆盖配置文件) export SANDBOX_PROBE_CONFIG="/path/to/config.yml"
"Do I trust this sandbox?" is a faith-based question — and faith is a rotten foundation for a threat model. Every AI coding agent ships with a story about what its sandbox can and cannot do: container policies, Landlock rules, seccomp filters, "we only allow reads from the workspace". That story is the vendor's map. You are defending the territory: a developer's laptop, with dotfiles, an SSH agent, cloud credentials, and an agent that any sufficiently clever prompt injection might persuade to go for a wander.
sandbox-probe is a single static Go binary you drop inside the sandbox — Claude Code, Gemini CLI, nono, a container, whatever — and let it look around. Run it once on the bare host for a baseline, then again inside the agent. The diff between the two reports is the sandbox boundary, measured rather than assumed. If it shows the agent can read ~/.aws/credentials, resolve arbitrary DNS, or reach 169.254.169.254, tighten the policy before you ship another line of code through it.
For building:
buf) — install via make install-buf (only required if you change the protobuf definitions)For end-to-end testing (depending on which sandboxes you want to exercise):
jq — JSON processor for parsing reportsdocker and/or podman — for containerised testingclaude-code — Claude Code CLI for Claude testinggemini-cli — Gemini CLI for Gemini testingnono — a Landlock/Seatbelt wrapper for AI agents and other programsmake build
git clone https://github.com/controlplaneio/sandbox-probe.git
cd sandbox-probe
make build
If you intend to run sandbox-probe inside a container, make sure it is built statically with standard library paths, or arrange for the relevant paths to be mounted in. This isn't usually an issue but can bite on non-glibc or non-FHS systems like Alpine, NixOS, or anything via Nix.
```bash
创新的AI安全工具,结合代理工作流与沙箱测试。虽然星数较少但方向前沿,适合AI安全领域专业人士。代码质量和文档完整度有待观察。
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
总体来看,沙箱探针 是一款质量良好的Agent工作流,在同类工具中具备一定竞争力。AI Skill Hub 将持续追踪其更新动态,建议收藏备用,结合自身场景选择合适时机引入使用。
| 原始名称 | sandbox-probe |
| 原始描述 | 开源AI工作流:Agentic sandbox enumeration: find security issues, and seed escape automation.。⭐22 · Go |
| Topics | AI安全代理工作流沙箱测试自主代理漏洞发现 |
| GitHub | https://github.com/controlplaneio/sandbox-probe |
| License | Apache-2.0 |
| 语言 | Go |
收录时间:2026-06-09 · 更新时间:2026-06-11 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端