Security audit
Security checks for vulnerabilities and agentic risk
This is a simple web-browsing instruction skill with some broad activation wording, but no hidden code, persistence, credential handling, or destructive behavior.
Install only if you want an agent to handle explicit web search, URL browsing, and page summarization requests. Be aware that its keyword wording is broad, so users should phrase browsing requests clearly and avoid giving credentials for login-only pages unless they trust the agent and site.
The trigger keywords include very common terms such as 'web', 'content', 'browse', and 'extract', making accidental or overly broad invocation likely. In an agent environment, ambiguous activation can route unrelated user requests into this skill, potentially causing unintended network access, URL fetching, or content handling without clear user intent.
The skill declares an MD+execute classification and says some functionality requires command-line execution capability, but the documented purpose is only web browsing, summarization, and content extraction. Expanding the expected execution surface to shell access creates unnecessary privilege and increases the chance that prompts or downstream instructions could trigger unsafe local command execution unrelated to browsing.
Repeating the same broad keyword list in the skill body reinforces ambiguous matching and still lacks clear boundaries for when the skill should activate. This increases the probability of unintended invocation and may expose the agent to unnecessary remote content retrieval or prompt-injection-bearing webpages during unrelated tasks.
The summary and description combine Chinese and English by default, and later sections continue mixing languages in the same skill document. This can violate language/locale policy expectations when the skill presents a language choice implicitly rather than based on user preference or documented regional scope.
