Security audit
Security checks for vulnerabilities and agentic risk
The skill is a petrochemical IT writing assistant, but it also explicitly teaches plagiarism-score reduction and AI-detection avoidance for papers.
Install only if you are comfortable with a skill that may help users bypass plagiarism and AI-authorship checks. The oil-and-gas IT and paper-formatting material is ordinary, but the evasion playbooks and under-specified scheduled crawling/push behavior should be removed or tightly restricted before normal use.
The documented capability to reduce plagiarism scores and evade AI-detection systems is unrelated to a legitimate industry IT expert function and directly facilitates academic integrity bypass. In context, these instructions operationalize deception by helping users submit disguised non-original or AI-generated work as authentic.
The skill provides concrete natural-language strategies for paraphrasing content to evade plagiarism checks and AI-detection systems. This goes beyond neutral editing support and constitutes actionable evasion guidance designed to defeat integrity controls, enabling fraud at scale.
The skill explicitly teaches how to make AI-generated text harder to detect by modifying style, entropy, and writing patterns. This is dangerous because it directly supports concealment of AI authorship in contexts where transparency and authenticity are required, such as school or professional evaluations.
The skill contains a superficial compliance disclaimer while explicitly enabling ghostwriting, plagiarism-evasion, and AI-detection evasion elsewhere in the file. This mismatch is dangerous because it can be used to legitimize deceptive academic misconduct and bypass policy review with misleading safety language.
The document is an explicit playbook for evading AI-text detection, including targeted advice on sentence entropy, perplexity-related signals, structural variation, and self-checks to improve bypass success. In the context of a skill that already advertises '过AI检测' and paper-writing support, this operationalizes deception and academic/procedural fraud rather than serving a legitimate petrochemical IT function.
The file directly instructs users on how to evade AI-detection systems by changing wording, structure, transitions, and industry-specific details to appear human-authored. Such guidance enables deliberate misrepresentation of authorship and undermines integrity controls used by schools, publishers, and other reviewers.
This content teaches users how detection systems work and how to modify output specifically to avoid being flagged, which is targeted circumvention guidance. Because the skill is presented as a petrochemical IT expert, this capability is unjustified by role and broadens the skill into facilitating misuse, including academic dishonesty and evasion of institutional controls.
The metadata frames the skill as an oil and gas IT expert, but the actual activation scope and behavior prominently include paper writing, rewriting, and evasion services. This deceptive positioning increases the chance that unsafe academic-fraud functionality is deployed under an unrelated professional label, reducing scrutiny and widening misuse.
The trigger list includes overly broad generic terms such as '论文', causing the skill to activate in many ordinary academic contexts without clear boundaries. Because the skill contains deceptive writing and evasion features, broad triggering increases exposure and the likelihood of misuse by users seeking prohibited assistance.
The workflow normalizes an end-to-end process for producing academic work and then refining it through '降重' and related post-processing, embedding deceptive assistance into routine operation. This structure makes misuse easier by turning misconduct into a guided service rather than an incidental response.
The skill describes scheduled scraping and push updates without clear user warning or consent controls around collection, retention, and outbound delivery. In a system handling user interests and uploaded materials, this can create privacy, compliance, and unauthorized-monitoring risks, especially if sources or delivery mechanisms are not transparent.
Even though the advice is framed as style improvement, it is clearly optimized to bypass AI-generated-text detection through paraphrased, non-keyword-based evasion techniques. This makes the content dangerous because it is harder to catch with simple filters and can be readily used to assist covert misuse at scale.
The document is entirely framed around Chinese-language paper formatting, including a required Chinese abstract and only optional English abstract. This imposes a specific language/locale preference in natural-language instructions without indicating user choice or opt-in.
