Security audit
Security checks for vulnerabilities and agentic risk
This storytelling coach is mostly coherent, but it deserves Review because it pushes users into sensitive emotional disclosure and reads or stores personal story material.
Install only if you are comfortable with a direct storytelling coach that may ask emotionally probing questions and use saved local story context. Before using it, confirm which `cerebro/` files it may read, decide whether completed stories should be saved, and stop or switch topics if the conversation feels too personal or distressing.
The activation criteria are extremely broad and include common conversational phrases like wanting help finding or telling a story. In an agent environment, this can cause the skill to trigger unintentionally during unrelated chats, leading to inappropriate behavior, context hijacking, or unintended access to other referenced resources and workflows.
The skill instructs the agent to read from and write user story data into persistent files under cerebro/, including saved histories, without a clear upfront user-facing disclosure about storage, retention, or sensitivity. Because the content being collected is personal and potentially emotionally sensitive, silent persistence increases privacy risk and the chance of retaining data the user did not expect to be stored.
The file explicitly instructs the agent to keep pushing deeper into vulnerable emotional material and frames discomfort as a sign of progress, while only minimally acknowledging safety boundaries. In a storytelling coaching skill, this can pressure users into disclosing trauma-adjacent or psychologically destabilizing content without adequate consent checks, distress screening, or a clear user-facing off-ramp.
The guidance hard-codes 988 for the United States and Canada as part of the required crisis response flow, which can be inappropriate or unusable for users outside those countries. In a crisis-handling skill, location-mismatched emergency guidance is safety-relevant because users may rely on the provided resource and lose time if it does not apply to them.
No suspicious patterns detected.
