Security audit
Security checks for vulnerabilities and agentic risk
This skill automates a disclosed CNKI paper-search and export workflow, with expected browser use and local result-file creation.
Install this only if you want an agent to operate CNKI in a browser and create exported result files locally. Before running it, confirm the keyword groups, expected export count, and save location, especially if the fallback Excel generation is used.
The trigger conditions are broad enough to activate on many ordinary research-related requests, which can cause the skill to run browser automation and downloading behavior without sufficiently specific user intent. In context, this increases the risk of overcollection, unintended site interaction, and unexpected local side effects from a casual query.
The skill performs downloads and local move/copy operations, including writing to ~/Downloads with generated filenames, without a clear upfront warning and explicit consent for filesystem modification. This is dangerous because it normalizes local file side effects and could surprise users or overwrite/duplicate files in ways they did not anticipate.
The skill instructs behavior beyond its declared scope by falling back to local Python/Excel generation when CNKI export is unavailable. That expands the tool from browser automation into local file creation and processing without an explicit manifest declaration or separate user consent, increasing the chance of unexpected file writes and unsafe execution paths.
The instructions state that execution must use the "openclaw" profile and must not use "chrome," presenting a fixed tooling/profile requirement without offering user opt-in or explaining a locale/policy-style organizational necessity. This is a natural-language constraint that removes user choice in a way that may conflict with policy expectations around opt-in for forced preferences.
Lines L052-L055 instruct the agent not to stop merely because CAPTCHA-related DOM text appears and to continue trying the form unless interaction is actually blocked. Later documentation says '验证码出现时必须请用户手动完成' and at L114 says to prompt the user if CAPTCHA appears, creating conflicting operational guidance about the same condition.
No suspicious patterns detected.
