Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed AI-text humanizer, but it explicitly optimizes for hiding AI authorship signals and can overwrite user files without confirmation.
Install only if you intentionally want a Chinese text-rewriting skill that changes style toward human-like prose. Avoid using it to misrepresent authorship, and do not give it file paths unless you are comfortable with the original file being replaced; a safer version should return rewritten text or require explicit confirmation and backups before writing.
The skill is explicitly designed to make AI-generated text appear human-written by removing detectable traits and injecting deceptive stylistic signals. In context, this is dangerous because it facilitates evasion of authorship disclosure, academic or editorial integrity checks, and downstream trust mechanisms that rely on honest provenance.
The multi-round rewrite loop systematically scores and intensifies techniques for human-authorship mimicry, including deliberate inconsistency, emotional injection, and anti-pattern suppression. This iterative optimization makes the deception more effective than ordinary style editing and increases the likelihood of misuse for fraud, plagiarism concealment, or policy evasion.
The skill tells the agent to read a user-supplied path and overwrite the same file without warning or confirmation. This is high risk because a single ambiguous prompt could lead to silent destructive changes to local content, and the user is not given an approval point to verify the target or the resulting edits.
The instruction '全程简体中文' mandates a specific language for all interactions. This is a locale/language restriction without any user opt-in or documented justification, which matches the policy-violation category for forced language behavior.
The manifest describes a rewriting skill, but the body extends behavior to reading files and overwriting the original path in place. This capability expansion is dangerous because users invoking a text-editing skill may not expect destructive file operations, increasing the chance of accidental data loss or misuse through path-based prompts.
The instruction to overwrite the provided file lacks a clear necessity boundary, safety checks, or backup flow. Even without malicious intent, direct overwrite creates a real risk of irrecoverable content corruption, especially when the skill's task can be completed by returning transformed text instead.
No suspicious patterns detected.
