Security audit
Security checks for vulnerabilities and agentic risk
This skill is a straightforward academic email lookup helper, but users should avoid overwriting their spreadsheet without a backup.
Before installing or using this skill, make a copy of the spreadsheet, confirm the exact input columns and output column, and review a sample of matches because same-name academics and stale public emails can produce wrong entries. Use it only for appropriate handling of publicly available contact information.
Overly broad trigger phrases can cause the skill to activate in routine conversations unrelated to the user's intent, leading to unintended bulk collection of personal contact data. In this skill's context, that is more sensitive because it automates scraping and writing email addresses into files, increasing privacy and misuse risk if triggered accidentally.
The skill describes writing results back into the original Excel file without a clear warning or confirmation step, which creates a risk of silent data loss or irreversible corruption of user data. Because the workflow is batch-oriented and automated, a mistaken run can alter many rows before the user notices.
The quick-start instructions explicitly direct the agent to overwrite the target Excel file with no safety prompt, backup, or dry-run option. In an automation context, this is dangerous because operator error, bad matches, or malformed data can cause immediate destructive changes at scale.
文件整体描述、触发词和操作说明均固定为中文语境，且没有说明可根据用户偏好切换语言。若该技能面向通用用户环境，这种默认单一语言设定可能构成语言/locale 选择不足。
The capability overview says the skill writes an email column into the original Excel file, but later usage examples and output sections claim a separate result file such as '/tmp/people_with_emails.xlsx'. These are materially different behaviors for file handling and output semantics, creating ambiguity about what the skill actually does.
The workflow states emails are identified specifically from search-result summaries, which implies a limited extraction scope. Later instructions direct the agent more broadly to run web_search and extract emails, including from academic platforms and PDFs, so the earlier wording does not match the broader actual documented behavior.
No suspicious patterns detected.
