邮件读取与管理技能 - 让 AI 能够读取、汇总、发送邮件。当用户要求查看邮件、汇总未读、发送邮件通知时触发此技能。

This skill appears to do what it claims (manage email via the himalaya CLI), but it has a few problems you should address before use: - Clarify dependencies: the registry should list 'himalaya' as a required binary so you know it must be installed. - Do NOT copy the example that puts passwords on the command line—this exposes credentials to shell history and process lists. Prefer OAuth or storing credentials in a secure environment variable or credential store. - If you enable scheduled reminders, confirm how the agent will store the schedule and whether it runs autonomously; be cautious granting ongoing autonomous access to your email. - When installing himalaya, use the official package source for your OS (brew, winget, crates.io) and verify the project repo. - Consider requiring explicit environment variables (or an OAuth flow) for credentials and avoid sharing raw auth tokens with the skill. If you need me to, I can suggest a safer configuration and a checklist to harden usage of this skill. Confidence is medium because the skill is instruction-only and coherent overall, but the missing metadata and insecure examples are concerning rather than definitive proof of malicious intent.

Type: OpenClaw Skill Name: email-reader Version: 1.0.0 The skill is designed for legitimate email management using the `himalaya` CLI tool. However, it instructs the AI agent to install software and execute shell commands (`himalaya` commands) based on user input. While the `SKILL.md` includes positive security recommendations (e.g., using environment variables for sensitive data), the inherent design of an AI agent executing unsanitized user input in shell commands creates a significant risk of shell injection (RCE vulnerability). There is no direct evidence of malicious intent like data exfiltration or persistence within the provided files, but the capability to execute arbitrary commands via injection makes it suspicious.