详细分析 ▾
ℹ 用途与能力The skill's name and description match what the SKILL.md instructs (finding suppliers, crafting messages, sending inquiries, checking replies). However, the runtime instructions require platform-specific tools (mcp__launchfast__supplier_research and mcp__claude-in-chrome__* chrome automation) and a logged-in Alibaba Chrome session, but these required tools are not declared in the skill's top-level metadata — a mismatch that should have been explicit.
⚠ 指令范围The instructions tell the agent to control the user's browser, navigate to supplier pages, read message pages, type and send messages, and take screenshots for verification. Taking screenshots of web pages opened in the user's browser can capture sensitive or unrelated information (account details, other messages). The use of coordinate-based clicks is brittle and can click the wrong UI element, causing unintended actions. While reading and sending messages is within the stated purpose, these operations are privacy- and action-sensitive and should be explicitly called out, limited, and audited.
✓ 安装机制No install spec or code files are present; the skill is instruction-only. This lowers supply-chain risk because nothing is downloaded or written to disk by the skill itself.
ℹ 凭证需求The skill requests no environment variables or credentials, and it does not ask for AWS/other unrelated secrets. Instead it relies on the user's logged-in Chrome session for Alibaba and on platform tools. That is reasonable for a browser-automation-based outreach skill, but the metadata should have declared these tool dependencies so users understand what platform capabilities it needs.
✓ 持久化与权限The skill does not request persistent 'always' inclusion or special agent-wide privileges. It can invoke autonomously (platform default), which is expected — but given the ability to operate the user's logged-in browser, users should be aware of the potential blast radius if the skill is allowed to run without supervision.