Token导航 LogoToken导航TokenDH.com
开发需要联网clawhub未标认证来源可访问clear审计通过

tool-output-auditor工具输出审核员

Agent Skill

用于辅助安全审计、权限检查、凭据风险、认证流程和常见漏洞排查。它适合让 Agent 梳理敏感配置、检查依赖风险、分析鉴权逻辑或生成安全复核清单。使用时不能把工具输出直接当最终结论,涉及密钥、令牌、用户数据或生产系统时,应先确认最小权限、脱敏方式和操作边界。

总安装

4,008

周安装

167

GitHub Stars

公开资料未说明

下载量

1,336
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:tool-output-auditor(工具输出审核员)
来源仓库:https://github.com/daowuu/tool-output-auditor
安装命令:
openclaw skills install tool-output-auditor
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install tool-output-auditor

简介

Tool-Output-Auditor 在代理执行命令前对工具输出进行审核与风险评估。

  • 适用于 OpenClaw 中需要保障多步骤工作流安全与可靠性的开发场景。
  • 核心能力包括权限校验、凭据风险识别与漏洞排查建议生成。
  • 通过 clawhub 安装后自动介入关键节点,输出审计结论。
  • 不能将审核结果直接作为最终决策依据,需人工复核高风险操作。

SKILL.md

name
tool-output-auditor
description
Audit tool and command results before acting on them. Use when an agent has already run a command or tool, especially in multi-step workflows where a failed, partial, or suspicious result could be accidentally ignored. Trigger for tasks involving validation, packaging, publishing, deployments, file edits, process management, API calls, or any workflow where the next action must depend on actual output rather than assumptions.
metadata
openclaw
requires
bins
[python3]
homepage
https://clawhub.ai

Tool Output Auditor

Always read the result before moving on. Treat tool output as evidence, not background noise.

Core rule

After a meaningful tool or command step, explicitly decide which of these states you are in:

  • success — output matches the intended action
  • failure — command or tool clearly failed
  • partial — some progress happened, but the intended outcome is not complete
  • ambiguous — output is incomplete, contradictory, or not enough to justify the next step

If the state is not clearly success, do not continue as if it succeeded.

Audit workflow

  1. Identify the exact step that just ran.
  2. Read the output, not just the exit status.
  3. Extract the decisive signals:

- exit code - explicit success/failure text - created/updated artifact - target version or ID - warnings that change the next action

  1. Classify the result as success, failure, partial, or ambiguous.
  2. Only then choose the next action.

What counts as decisive evidence

Good evidence

  • Successfully packaged skill to ...
  • Published <slug>@<version>
  • Latest: 0.1.0
  • file exists at the expected path
  • expected process state is visible

Bad evidence

  • command started but did not finish
  • logs were truncated before the important line
  • output contains an error but the agent keeps going
  • a different command succeeded, but the target command failed
  • a result was assumed from memory instead of verified now

Common anti-patterns

Ignoring explicit errors

If output says Validation failed, Path must be a folder, 401, or Skill not found, stop and address that exact error.

Mistaking activity for success

Preparing..., Fetching..., or a running background job is not success.

Checking the wrong thing

Do not treat a successful pre-step as proof that the real step succeeded.

Narrating future intent as if it already happened

Do not say "published successfully" until the output proves publication.

Recommended use

Use scripts/audit_output.py when you want a quick second-pass summary of a captured command result. It is especially helpful for:

  • package / publish workflows
  • deployment commands
  • API auth failures
  • background process completion
  • file-generation steps

Scripts

scripts/audit_output.py

Classifies a saved output file or stdin text into success, failure, partial, or ambiguous, highlights the decisive lines, detects contexts such as publish/package/auth/process, and gives a next-step recommendation.

References

  • Read references/checklist.md before building automation around multi-step command workflows.
  • Read references/examples.md for concrete examples of success, failure, partial, and ambiguous outputs.
  • Read references/real-mistakes.md for real failure patterns this skill is meant to prevent.

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

82.98%
按下载量换算1,109

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills