← 返回 Skills 市场
155
总下载
0
收藏
0
当前安装
2
版本数
在 OpenClaw 中安装
/install reikys-hallucination-guard
功能描述
Execution-based verification guardrail with 14 check items for AI agent output
安全使用建议
This skill appears to do what it says: a checklist of concrete commands to verify agent outputs. Before using it, decide where the checks will run — locally, in a sandbox, or on a remote runner — because the scripts read local paths and may query external URLs. Don’t blindly add the auto-execution snippet to your system prompt unless you want these checks to run on every agent task; consider running on-demand or restricting scope to non-sensitive files. Verify presence and safety of invoked tools (curl, python, node, jq, etc.) in your environment, and inspect any output before sharing it externally. If you need higher assurance, run the checks in an isolated environment (container/VM) or review/modify the SKILL.md to remove checks you don’t want performed.
能力评估
Purpose & Capability
The name/description (execution-based verification with 14 checks) matches the SKILL.md contents: the file contains concrete, actionable checks (file existence, binary presence, URL checks, syntax checks, numerical sourcing, consistency checks, etc.). The skill requests no credentials or installs, which is reasonable for an instruction-only verification tool.
Instruction Scope
SKILL.md explicitly instructs running shell commands (stat, ls, command -v, curl, python/node/jq, bash -n, etc.) against local paths and external URLs. That behavior is coherent for verification, but the instructions access local filesystem state and make network requests — both are privacy-sensitive. The guide also suggests auto-execution in the system prompt, which would cause these checks to run frequently unless the user limits scope.
Install Mechanism
No install spec and no bundled code — lowest-risk delivery model. The skill relies on standard system tools being present but does not download or install anything.
Credentials
The skill requests no environment variables, credentials, or config paths. Its checks reference common binaries and network endpoints but do not require secrets; this is proportionate to the stated purpose.
Persistence & Privilege
Skill metadata does not force persistence (always: false) and allows normal autonomous invocation. The SKILL.md suggests adding the guard to the system prompt for auto-execution; that is a user action (not enforced by the skill) but would increase runtime frequency and blast radius if adopted — users should be cautious about automatic invocation on every task.
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 reikys-hallucination-guard - 安装完成后，直接呼叫该 Skill 的名称或使用
/reikys-hallucination-guard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
v1.0.1
Translated all content to English
v1.0.0
hallucination-guard v1.0.0
- Initial release.
- Provides 14 execution-based checklists to validate AI agent outputs for hallucinations in areas such as paths, commands, library/package names, numeric claims, completeness, consistency, and deliverable existence.
- Offers quick and full check modes, with specific shell and PowerShell commands for each validation category.
- Reports results in structured PASS/FAIL format, enabling actionable, reliable AI output verification.
- Documentation available in Korean, with detailed sample commands and criteria for each check.
元数据
常见问题
Hallucination Guard 是什么？
Execution-based verification guardrail with 14 check items for AI agent output. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 155 次。
如何安装 Hallucination Guard？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install reikys-hallucination-guard」即可一键安装，无需额外配置。
Hallucination Guard 是免费的吗？
是的，Hallucination Guard 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Hallucination Guard 支持哪些平台？
Hallucination Guard 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 Hallucination Guard？
由 reikys（@reikys）开发并维护，当前版本 v1.0.1。
推荐 Skills