Security audit
Security checks for vulnerabilities and agentic risk
The skill is a coherent patent-mining assistant, but it requires sensitive project context to be copied verbatim into final reports without minimization or redaction.
Install only if users understand that patent project details, company context, competitor notes, constraints, and other CP-0 additions may be copied verbatim into generated reports. Avoid entering confidential details unless they are intended to appear in the final report, and review generated reports before sharing them.
The file states that all user-facing outputs must follow language rules expressed as mandatory requirements and the entire skill description is written to enforce Chinese technical phrasing, with no opt-in or alternative locale choice. This creates a language/locale policy issue because it effectively forces one language style on all users regardless of preference or context.
This instruction requires confirmed user-supplied information to be written verbatim into the final report, including potentially sensitive business context, constraints, or identifiers. That creates a semantic data-leak risk because private inputs gathered during confirmation are transformed into persistent output without minimization or sensitivity filtering.
The skill explicitly instructs the agent to internally retain a complete CP-0 snapshot and later reproduce all additional user information in the report without summarization. This is dangerous because it formalizes collection, retention, and replay of arbitrary user text, increasing the chance that confidential business data, personal details, or sensitive competitive information will be disclosed downstream.
The final report specification mandates verbatim inclusion of user supplementary information and forbids omission, replacement, or summarization. In a patent-mining context, users are likely to disclose non-public R&D direction, competitive assumptions, or company background, so this rule materially increases the likelihood of confidential information appearing in a sharable artifact.
This quality rule reinforces mandatory retention and complete write-back of confirmation snapshot fields, making the leakage behavior systemic rather than incidental. Repetition across the skill increases the chance an implementation will treat full user replay as required behavior, causing unnecessary exposure of sensitive user input in generated reports.
No suspicious patterns detected.
