Back to skill

Security audit

邮件摘要

Security checks for vulnerabilities and agentic risk

Overview

This email-summary skill is mostly coherent, but it asks to automate sensitive webmail sessions while also advertising anti-crawler bypass and persistent local/email snapshot storage.

Install only if you are comfortable letting an agent use your logged-in browser session to read mailbox metadata and create local reports/screenshots. Avoid any anti-bot or anti-crawler bypass use, choose a private output directory, limit retention, and review any scheduled cron or launchd job before enabling it.

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (3)

Intent-Code Divergence

High
Confidence
96% confidence
Finding
The skill explicitly advertises '反爬虫策略自动绕过' while elsewhere claiming verification and CAPTCHA cannot be bypassed. This contradiction is dangerous because it normalizes anti-detection and anti-abuse behavior for a tool that automates access to highly sensitive email accounts, increasing the risk of stealthy unauthorized access or policy-violating automation.

Context-Inappropriate Capability

High
Confidence
97% confidence
Finding
Advertising anti-crawler bypass is unjustified for a daily email summary skill and materially expands it from benign automation into evasive access tooling. In the context of webmail automation, such capability can be abused to defeat provider defenses and facilitate covert collection of mailbox metadata or content.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill writes screenshots and Markdown reports locally, containing email senders, subjects, timestamps, and potentially inbox snapshots, but does not prominently warn users about local persistence of sensitive metadata. This can lead to unintended exposure through shared machines, backups, logs, or weak filesystem permissions.

Static analysis

No suspicious patterns detected.