Security audit
Security checks for vulnerabilities and agentic risk
This skill is a LinkedIn reply assistant that can publish a live reply and reaction after approval, with no hidden scripts or persistence found.
Before installing, understand that approving a draft is intended to modify your LinkedIn presence by sending both a reaction and a reply through Publora. Review the exact reply text, reaction type, target comment URL, and parentComment before approving, especially for business or public-facing accounts.
The phrase 'Use when the user wants to reply to a comment on any post' is broad and could overlap with many ordinary LinkedIn-reply requests without clearly constraining when this skill should be chosen versus related skills. Although the line excludes top-level comments, it does not provide explicit negative examples or tighter trigger boundaries beyond that.
The skill states that an approval card leads to a live reaction and reply via Publora, but it does not prominently warn the user that approval triggers external side effects on their LinkedIn presence. In an agent setting, weak disclosure around real-world posting can cause unintended account actions, reputational harm, or user surprise if the approval signal is ambiguous or auto-forwarded.
The example depicts a real external side effect sequence: reacting on LinkedIn, waiting, and then posting a reply, but it does not show any explicit warning that these actions will transmit data to third-party services and modify the user's LinkedIn presence. In a skill whose purpose is to draft and post replies via Publora, omission of clear user-facing confirmation around each outbound action increases the risk of unintended posting or social-account activity being normalized as automatic behavior.
The documentation introduces a behavioral requirement to always add a reaction when replying, even though the skill is described as drafting and posting a specific LinkedIn reply. This expands the action scope from a single approved reply into an additional engagement action the user may not have requested, creating a risk of unauthorized or misleading account activity and violating least-privilege/user-intent boundaries.
The instruction "No em dashes. Use .. for soft pauses." imposes a specific writing style globally in natural language. This can be treated as a policy-style constraint on output formatting without indicating that the user may choose or override it.
No suspicious patterns detected.
