Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed Chinese-language knowledge-base distillation skill that reads user-specified content to generate installable skills, with no evidence of hidden code, exfiltration, or persistence beyond normal skill installation.
Install this if you want a Chinese-language workflow for turning selected knowledge bases into reusable skills. Use explicit prompts with the knowledge-base name and scope, avoid giving it private or copyrighted material you do not want summarized into a skill, and review any generated Skill before installing or sharing it.
The README is entirely written in Chinese and presents all invocation and usage instructions in that language, without offering a language choice or stating that the skill is intentionally region- or language-specific. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern.
The README instructs users to trigger the skill with very broad natural-language phrases such as “盘古” or “蒸馏知识库”. In a chat-based agent environment, such generic phrases can be mentioned incidentally or in quoted text, causing unintended activation and unauthorized reading or transformation of a knowledge base.
The installation and usage section describes activation in ambiguous terms and does not specify clear conditions for when the skill should run. In multi-agent or assistant contexts, this increases the chance that normal conversation, copied examples, or third-party content will accidentally invoke the skill and start a multi-phase workflow with data access implications.
The trigger list includes broad and ambiguous activators such as the single-word phrase "盘古" alongside several generic distillation phrases. In a chat environment, this raises the likelihood of accidental or contextually inappropriate activation, which could cause the skill to read or transform knowledge base content when the user did not intend to invoke the full workflow.
The usage guidance explicitly tells users to activate the skill by saying only "盘古" or "蒸馏知识库," which does not establish clear invocation boundaries. Because this skill can initiate multi-phase content reading and skill generation, vague activation increases the risk of unintended execution, unnecessary access to user content, and confusing agent behavior.
The embedded SKILL.md states '本 Skill 只做中文合同审查', which imposes a language/locale restriction in natural-language instructions. The file does not offer the user a language choice or provide a clearly documented compliance or regional justification for restricting the skill to Chinese contracts only.
The manifest description and trigger list include generic phrases such as “日期公式”, “逻辑公式”, “日期函数”, and “逻辑函数”, which could match common user requests rather than a narrowly scoped skill invocation. Although the file includes some non-applicable scenarios later, it does not clearly constrain activation conditions or provide explicit negative trigger examples for these phrases.
This markdown file is entirely written as a Chinese-only skill example, including user inputs, triggers, and skill outputs, with no indication that users may choose another language. Under the policy criteria, forcing a specific language without opt-in is a natural-language policy violation unless the locale restriction is explicitly justified.
This markdown file contains user-facing instructional content exclusively in Chinese, and nowhere indicates that the skill is region-specific or that users may choose another language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.
