Security audit
Security checks for vulnerabilities and agentic risk
This is a Markdown-only text extraction skill whose privacy-sensitive behavior is disclosed and aligned with its purpose.
Install only if you are comfortable with the agent consolidating conversation or document content into reusable Q&A and preference summaries. Avoid invoking it on chats or documents containing sensitive personal, business, legal, medical, or credential information unless you explicitly want that material summarized.
The skill’s trigger conditions are broad enough to activate on common summarization, note-taking, or recap requests, which can cause the agent to perform full-session extraction when the user did not explicitly ask for structured Q&A or preference mining. This increases the chance of over-collection and unintended disclosure of sensitive conversational content.
The skill encourages scanning full chat sessions and uploaded documents without warning that it may surface sensitive personal data, decisions, or inferred preferences. Users may unknowingly cause the agent to aggregate and restate information that is more revealing in structured form than in the original conversation.
The skill is explicitly designed to mine and restate user conversation contents and preferences, which creates a privacy risk because it transforms dispersed chat history into a compact, reusable knowledge artifact. That restructuring can expose sensitive attributes, choices, or contextual details that were not originally intended to be consolidated.
Requiring the model to scan the full conversation from the first message and infer preferences broadens collection beyond the immediate request and increases the risk of pulling in irrelevant or sensitive historical context. The instruction to derive implied preferences is particularly risky because it encourages inference about user traits or intentions that may be inaccurate or privacy-invasive.
No suspicious patterns detected.
