← 返回 Skills 市场
96
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install robust-text-auditor
功能描述
(已验证) 高可靠性文本审核器，结合本地规则、百度 API 和 AI 深度分析，提供三层合规性审查。
安全使用建议
Before installing or running this skill, do the following checks: 1) Verify the missing orchestration script: SKILL.md references run.ps1 but the package contains only Python scripts; ask the author for the actual entrypoint and a clear invocation example. 2) Inspect baidu_api.py yourself: it sends data to official Baidu endpoints (aip.baidubce.com) — expected, but confirm you are comfortable sending audit text to that cloud. 3) Do NOT blindly paste your global credentials into ~/.openclaw/config.json — prefer creating a limited-scope Baidu key or using a per-skill config file. The skill metadata should declare the required config path and credential needs; request that from the publisher. 4) Ensure the skill’s output path and temporary files are acceptable (it writes prompt/output files in the skill directory). 5) If you will run this in automated/autonomous mode, remember the agent may invoke the skill and send text to Baidu without additional prompts — restrict usage to non-sensitive test data until you trust it. If the author cannot explain the run.ps1 discrepancy and update the metadata to declare the config requirement, treat the package as untrusted.
功能分析
Type: OpenClaw Skill
Name: robust-text-auditor
Version: 1.0.1
The 'robust-text-auditor' skill is a legitimate tool designed for multi-layered text compliance auditing. It utilizes a local keyword scanner (local_scanner.py), integrates with official Baidu Cloud APIs for cloud-based censorship (baidu_api.py), and generates a structured prompt for further LLM analysis (create_final_prompt.py). The code follows standard practices for OpenClaw skills, such as reading credentials from the user's home directory (~/.openclaw/config.json) and communicating only with official endpoints (aip.baidubce.com). No evidence of data exfiltration, malicious execution, or prompt injection against the agent was found.
能力评估
Purpose & Capability
The declared purpose (three-layer auditing using local rules, Baidu API, and AI) matches the included code: local_scanner.py (rules.json), baidu_api.py (calls Baidu endpoints), and create_final_prompt.py (builds the LLM prompt). However the SKILL metadata did not declare that Baidu credentials or a config path are required, even though the code clearly needs credentials in ~/.openclaw/config.json. That omission is inconsistent (not necessarily malicious) but important.
Instruction Scope
SKILL.md repeatedly instructs the agent to run a core run.ps1 orchestration script and to add Baidu credentials to ~/.openclaw/config.json. There is no run.ps1 included in the file manifest (only Python scripts). The instructions therefore do not match the shipped files and give ambiguous guidance about how the agent should invoke the skill. Also SKILL.md instructs the user to place credentials into a global config path (~/.openclaw/config.json) which the provided baidu_api.py will read; this is a sensitive operation and it was not declared in the skill metadata.
Install Mechanism
No install spec is provided (instruction-only), and all code shipped is plain Python source — there is no downloaded binary or external installer. This is lower risk than arbitrary downloads, but the repository includes executable scripts that will be run; the lack of an install step means the agent or user must execute them directly. The code imports requests (declared in SKILL.md requirements), which is reasonable for network calls.
Credentials
The skill requires Baidu API credentials, but the registry metadata lists no required env vars or config paths. The SKILL.md instructs the user to add baidu_api_key and baidu_secret_key to ~/.openclaw/config.json; baidu_api.py indeed reads that file. Requesting global config credentials is proportionate to calling Baidu's API, but the omission from metadata and the instruction to store credentials in a shared user config file is a security/privacy concern and should have been declared explicitly.
Persistence & Privilege
The skill does not request permanent inclusion (always:false) and does not declare any special system privileges. It will read the user's ~/.openclaw/config.json and write output files (create_final_prompt.py writes to an output path; SKILL.md says final_prompt.txt will be created in the skill directory). Reading a home-config file is a modest privilege — acceptable for a service needing API keys — but combined with the metadata omissions it raises caution.
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 robust-text-auditor - 安装完成后，直接呼叫该 Skill 的名称或使用
/robust-text-auditor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
- Updated SKILL.md with detailed usage instructions, prerequisites, and output information.
- Clarified the multi-layered auditing approach: local rule scanning, Baidu API review, and AI-based deep analysis.
- Specified required configuration of Baidu Cloud API credentials.
- Provided invocation examples for both file-based and string-based auditing.
- Described output format and expected workflow for integrating with language models.
元数据
常见问题
高可靠性文本审核器 是什么？
(已验证) 高可靠性文本审核器，结合本地规则、百度 API 和 AI 深度分析，提供三层合规性审查。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 96 次。
如何安装 高可靠性文本审核器？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install robust-text-auditor」即可一键安装，无需额外配置。
高可靠性文本审核器 是免费的吗？
是的，高可靠性文本审核器 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
高可靠性文本审核器 支持哪些平台？
高可靠性文本审核器 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 高可靠性文本审核器？
由 an（@ahsbnb）开发并维护，当前版本 v1.0.1。
推荐 Skills