Security audit
Security checks for vulnerabilities and agentic risk
This skill is a Chinese WeChat-style rewriting assistant, but it includes instructions to disguise AI-written text as human and contains example workflows for scraping third-party content.
Review carefully before installing. The skill does not appear to install code or gain persistence, but users should avoid using it to hide AI authorship, impersonate real authors, scrape third-party platforms, or paste real API keys into chat or generated files.
The skill claims applicability to broad scenarios like '风格模仿', '文章改写', and '公众号改写' without clear boundaries or disambiguation rules. This ambiguity can cause accidental invocation and expands exposure to the skill's more problematic behaviors, including author-style imitation and humanization tactics.
The skill claims applicability to broad scenarios like '风格模仿', '文章改写', and '公众号改写' without clear boundaries or disambiguation rules. This ambiguity can cause accidental invocation and expands exposure to the skill's more problematic behaviors, including author-style imitation and humanization tactics.
The skill explicitly instructs the model to insert deliberate spelling and punctuation mistakes so AI-generated content appears human-authored. That is deceptive provenance laundering: it is designed to conceal machine generation from readers, reviewers, or platform detection systems, and can facilitate impersonation or policy evasion.
The combined instructions to add natural-looking errors and then perform a silent self-check to ensure the output no longer has obvious 'AI味' amount to a workflow for stealthily disguising AI-generated text as human writing. The hidden optimization step makes the deception more robust and intentional, increasing misuse risk for ghostwriting, impersonation, and bypass of disclosure or moderation expectations.
This markdown file presents all instructions exclusively in Chinese and does not indicate that the skill is region-specific or that users may choose another language. Under the policy, forcing a specific language without opt-in can be a natural-language policy violation.
This file prescribes a single language and locale for the skill content through its natural-language instructions and examples. Under the policy, forcing a specific language without user opt-in can be a language/locale policy violation unless the constraint is explicitly justified.
The style reference embeds concrete instructions to log into Xiaohongshu, simulate human browsing, extract third-party posts, and save the content into files. In a writing-style skill, this normalizes unauthorized scraping and collection of third-party content without any guardrails about terms of service, copyright, consent, or privacy, making downstream misuse more likely.
The sequence instructs the system to open a website, wait for user login, mimic human behavior, collect posts meeting engagement thresholds, and save the results. That combination goes beyond harmless style description and provides an operational pattern for stealthy bulk harvesting of third-party content, which is especially risky because this skill is meant to rewrite content in others' styles.
The document tells users to set an environment variable with an API key placeholder but provides no warning that the value is a secret requiring secure handling. In skill content, this can encourage users or agents to paste credentials into prompts, logs, or unsafe configuration paths, increasing risk of secret exposure.
This markdown file is entirely framed as a Chinese writing-style profile, with all examples, structural guidance, and fixed closing text specified only in Chinese. Because the skill content appears to prescribe a single language/locale for outputs and does not mention user opt-in or a region-specific justification, it conflicts with the policy against forcing a specific language without user choice.
No suspicious patterns detected.
