Security audit
Security checks for vulnerabilities and agentic risk
This skill is for automated X/Twitter engagement, but it gives an agent authority to post public replies from a logged-in account without confirmation and includes ambiguous like/follow-up behavior.
Install only if you are comfortable with an agent using a logged-in X/Twitter profile to choose posts and publish replies automatically. Review the like-action ambiguity and disable or tightly control the optional follow-up tracking unless you explicitly want persistent engagement history and revisits.
The skill explicitly mandates fully automated commenting and direct sending without asking for confirmation, which is dangerous because it performs irreversible public posting actions on a user's social account. In context, this increases the chance of reputational harm, policy violations, or posting unwanted content if topic selection or generated text is mistaken.
The inline documentation is self-contradictory about whether the skill likes tweets, which creates ambiguity in a high-risk domain involving authenticated account actions. Ambiguous execution docs can cause an agent to select the more permissive path and perform actions the user did not intend, especially when paired with the skill's default auto-send behavior.
The skill positioning section explicitly describes the skill as automatically finding content, liking it, composing a reply, and sending it, while the manifest says comment-only. This kind of scope contradiction is dangerous because downstream agents may follow the broader behavioral description and perform unintended social actions on a logged-in user account without clear consent.
该技能在描述中用“去 X 时间线上找几条技术帖子评论”“帮我在 X 推荐页发几条技术评论”等自然语言作为触发示例，但没有给出明确的触发边界、排除条件或固定调用短语。对于 markdown 文件，这会造成何时应自动调用该 skill 的范围不清，增加意外触发的风险。
The manifest repeatedly states the skill should only publish comments and must not perform likes. However, the documented implemented capabilities explicitly say it can click like/reply buttons, which broadens the described interaction behavior beyond comment-only operation.
文档规定“中文推文→中文评论，其他语言→英文评论”，属于预设的语言/locale 策略。根据规则，若技能强制特定语言而未提供用户选择或 opt-in，则构成自然语言政策问题。
The optional tracking feature stores persistent interaction history and drives future follow-up behavior via cron-like revisits, extending the skill beyond one-off commenting into ongoing profiling and automated engagement. This is risky because it creates durable behavioral records tied to other users and can lead to repeated autonomous actions without fresh user awareness or consent.
The workspace persistence of interaction records is implemented without any warning, consent, retention limit, or privacy controls. That is dangerous because it quietly creates a local dossier of social interactions and enables future automated engagement based on stored behavioral history.
No suspicious patterns detected.
