← 返回 Skills 市场
MeatLoop
作者
meatloopservice
· GitHub ↗
· v1.0.0
· MIT-0
305
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install meatloop
功能描述
A real human reviews your question, screenshot, or pair of images and returns a structured verdict by email. Use for sanity checks, content verification, cre...
安全使用建议
Before installing or using this skill consider: (1) It asks you to send images/questions to a third‑party Gmail address — do not send sensitive or confidential screenshots (credentials, financial, medical, legal) to that address. (2) The skill provides no declared mechanism for the agent to send/receive email automatically; ask the author how email is intended to be sent and how replies are ingested. (3) Ask for a privacy/data-retention policy and details on the public verification log (what is posted, could it leak identifiers?). (4) If you must test, use throwaway accounts and anonymized/redacted data. (5) Prefer a skill that declares required credentials/tools (SMTP/API keys) and documents how responses are delivered and data is protected — absence of those is the primary incoherence here.
功能分析
Type: OpenClaw Skill
Name: meatloop
Version: 1.0.0
The 'meatloop' skill (SKILL.md) instructs the AI agent to send potentially sensitive data, such as screenshots of deployment dashboards, invoices, and login pages, to an external third-party email address ([email protected]) for 'human review.' While framed as a human-in-the-loop utility, this design facilitates the exfiltration of private or internal information to an unauthenticated destination. The use of a generic Gmail account and a public Google Spreadsheet for 'verification' (IOC: 2PACX-1vTNynmFGYxtUetqMgvGsO4VY6TE_i-ZDdotEFweE_9QsZo4njPpBhrHZ5aYbTC7Ql-8GnwgN2NHnHXi) are significant red flags for a data harvesting operation disguised as a legitimate service.
能力评估
Purpose & Capability
The stated purpose (human-in-the-loop image/question review) is plausible and matches the SKILL.md content. However, the skill is presented as an agent skill but declares no way for the agent to send or receive the required email (no SMTP/API credentials or tool bindings). That mismatch may mean the skill expects a human to manually email the service, or it omits required environment/tool declarations.
Instruction Scope
Runtime instructions tell the agent/user to send potentially sensitive images and questions to [email protected] and expect a reply by email; they also state a tamper-evident verification code will be posted to a public log. The instructions therefore direct data to an external endpoint (a third‑party Gmail account and a public log) and do not describe how replies are received back into the agent, how long data is retained, or what metadata is published — all of which are meaningful privacy/exfiltration concerns.
Install Mechanism
No install spec and no code files are present (instruction-only). This is low‑risk from a code execution/install perspective because nothing will be written to disk by the skill itself.
Credentials
The skill declares no required environment variables or credentials despite requiring sending email and receiving email responses. If the agent is expected to send email automatically, SMTP or mailbox credentials would be required but are not declared. Separately, the instructions encourage uploading images/text to an external Gmail account — a proportionality/privacy concern because that can leak PII or sensitive screenshots.
Persistence & Privilege
The skill does not request persistent system privileges and always:false. Autonomous invocation is allowed (default) — combined with the external-email data flow this increases potential exposure if the agent automatically emails content to the third party. The skill does not modify other skills or system settings.
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 meatloop - 安装完成后，直接呼叫该 Skill 的名称或使用
/meatloop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 Meatloop skill.
- Provides human-in-the-loop verification and judgment for questions and images via email.
- Supports binary and near-binary questions, content and UI verification, creative/tone checks, document and receipt inspection, and A/B image comparison.
- Structured verdicts with human-generated VERIFICATION codes for each substantive response.
- Simple submission: send an email (optionally with up to 2 images); no setup or credentials required.
- Ensures privacy: attachments deleted after review, only minimal metadata retained.
元数据
常见问题
MeatLoop 是什么？
A real human reviews your question, screenshot, or pair of images and returns a structured verdict by email. Use for sanity checks, content verification, cre... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 305 次。
如何安装 MeatLoop？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install meatloop」即可一键安装，无需额外配置。
MeatLoop 是免费的吗？
是的，MeatLoop 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
MeatLoop 支持哪些平台？
MeatLoop 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 MeatLoop？
由 meatloopservice（@meatloopservice）开发并维护，当前版本 v1.0.0。
推荐 Skills