Security audit
Security checks for vulnerabilities and agentic risk
This skill is clearly meant to publish a single X post, but it can take the irreversible step of posting publicly without requiring final user approval of the exact text.
Review this skill carefully before installing. It is suitable only if you want an agent to publish live X posts from your account. Ask for draft-only behavior or require explicit confirmation of the exact final tweet and account before the agent clicks publish.
The workflow proceeds to compose and publish a live X post without an explicit user-facing warning in the skill description that account content will actually be posted. This is especially dangerous here because the skill includes browser-driven execution against x.com, so a user seeking help with wording or research could unintentionally trigger a public post on their account.
The skill description is broad enough to match ordinary requests about researching a topic, drafting a tweet, or posting on X, which increases the chance the agent invokes a live-posting workflow when the user did not clearly intend account actions. In this context, overbroad triggering is more dangerous because the skill culminates in publishing to a real social media account, creating risk of unintended public actions rather than a harmless drafting assist.
The file says to use it whenever drafting a comment, reply, or post, which creates a broad default activation rule detached from the specific skill scope. In a skill that researches topics and posts to X in the user's voice, this can cause the persona/style prompt to be applied to ordinary writing tasks without explicit user consent, increasing the chance of overbroad behavioral steering and unintended impersonation-like output.
The file imposes unconditional bans on broad word classes and framing patterns, overriding normal user-preference flexibility and steering output style in a rigid way. In this skill, that rigidity is less about direct security compromise and more about covert policy shaping: it can suppress user-intended tone, distort faithful assistance, and strengthen deceptive 'humanized' impersonation behavior when posting publicly in the user's voice.
The workflow directs the agent to publish a post on X as part of its normal sequence, but it does not require an explicit final user confirmation immediately before the irreversible external action. That creates a real risk of unintended posting if the topic, draft, or account context is misunderstood, especially because the skill is designed for live browser execution on a real social account.
No suspicious patterns detected.
