← 返回 Skills 市场
106
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install safe-fuzzer-skill
功能描述
Sandbox-only behavior-led gray-box skill fuzzer. Spawns a worker subagent, probes an installed target skill, deploys honeypot fixtures, and returns a structu...
安全使用建议
This skill is broadly coherent with its stated purpose, but do not run it on a non-isolated environment. Only run from a locked sandbox with verified host boundary enforcement and elevated exec disabled. Before installing/using: (1) verify your sandbox maps workspaces so that any recommended host paths (e.g., ~/.openclaw/workspace-fuzzer) are inside the sandbox or change them to sandbox-local paths, (2) run with the smallest preset first (min) and synthetic fixtures only, (3) confirm the platform enforces the SKILL.md preflight checks (sandbox presence and no elevated exec), and (4) review the target skill's repository for embedded secrets or host-path references because the fuzzer is allowed to read target-owned files during gray-box planning. If you cannot guarantee isolation or you see host-paths being used, treat this skill as high-risk and do not run it against real or sensitive workloads.
功能分析
Type: OpenClaw Skill
Name: safe-fuzzer-skill
Version: 1.0.0
The safe-fuzzer skill is a security testing tool designed to perform gray-box fuzzing on other OpenClaw skills by spawning worker subagents and monitoring their behavior. It employs high-risk capabilities, including shell execution, network egress probing, and sub-session management (SKILL.md), which are plausibly necessary for its stated purpose but inherently risky. The skill includes robust safety gates, such as mandatory sandboxing and the use of synthetic 'honeypot' secrets (balanced.json), and explicitly prohibits accessing real host environment variables or sensitive OpenClaw configuration files.
能力评估
Purpose & Capability
Name/description match what the SKILL.md instructs: spawn a worker subagent, deploy honeypot fixtures, probe a target skill, and produce a structured JSON report. Use of sessions_spawn/sessions_send to create a worker subagent is expected for this purpose.
Instruction Scope
Instructions explicitly require sandboxed execution and forbid reading host auth files and env vars; they allow limited gray-box reads of target-owned files (./skills/<target>/**) for planning. This scope is reasonable for a gray-box fuzzer, but it grants the agent discretionary read access to target source/docs which could include sensitive data. The SKILL.md relies on the runtime to enforce the preflight checks (sandbox present, elevated exec unavailable)—the instructions cannot enforce that themselves.
Install Mechanism
Instruction-only skill with no install steps or remote downloads. No packages or external installers are pulled by the skill files included—this lowers install-time risk.
Credentials
The skill requests no environment variables or credentials, which is proportional. However README guidance suggests installing the tested target into ~/.openclaw/workspace-fuzzer; that host-path recommendation could cause users to place target artifacts in the host home directory outside the sandbox unless they understand how their sandbox maps home/workspaces. Confirming sandbox isolation is therefore essential.
Persistence & Privilege
always:false and disable-model-invocation:true (no autonomous model invocation) reduce persistent/automatic risk. The skill spawns worker subagents for execution, which is coherent with its purpose; this increases blast radius if the runtime's session-spawn API isn't properly sandboxed, but the skill itself does not request always-on privileges or cross-skill config changes.
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 safe-fuzzer-skill - 安装完成后，直接呼叫该 Skill 的名称或使用
/safe-fuzzer-skill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 release of safe-fuzzer.
- Provides a gray-box, behavior-led fuzzer for installed skills, operating only inside a locked sandbox.
- Orchestrates fuzzing using a parent/worker agent model, with parent setup and worker executing probes.
- Supports configurable presets for probe depth (`min`, `balanced`, `max`), with CLI invocation and timeout recommendations.
- Deploys honeypot fixtures, probes skill execution, records file/shell/network behavior, and returns a structured JSON risk report.
- Includes strict safety gates to prevent host compromise, secrets leakage, or running outside the sandbox.
- Comprehensive execution flow: preflight, target resolution, test strategy planning, fixture deployment, probe loop, and final reporting.
元数据
常见问题
SAFE-Fuzzer 是什么？
Sandbox-only behavior-led gray-box skill fuzzer. Spawns a worker subagent, probes an installed target skill, deploys honeypot fixtures, and returns a structu... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 106 次。
如何安装 SAFE-Fuzzer？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install safe-fuzzer-skill」即可一键安装，无需额外配置。
SAFE-Fuzzer 是免费的吗？
是的，SAFE-Fuzzer 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
SAFE-Fuzzer 支持哪些平台？
SAFE-Fuzzer 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 SAFE-Fuzzer？
由 Agentsey（@agentsey）开发并维护，当前版本 v1.0.0。
推荐 Skills