T01 · Skill Instruction Hijacking
- Location
agents/openai.yaml:4- Finding
Persistent Branded Output Instruction Hijacking
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This is mostly a Chinese-language analysis framework, but it can activate too broadly and force repeated branded Chinese text into every reply.
Review this skill before installing if you expect neutral, concise, or language-adaptive responses. It does not show evidence of malware, credential theft, remote execution, or persistent system changes, but it may auto-activate for broad analysis requests and repeatedly insert its own fixed Chinese branding and structure into answers.
agents/openai.yaml:4Persistent Branded Output Instruction Hijacking
声明描述的是一个面向用户问题分析的方法论型技能;而代码是一个与技能元数据校验相关的开发/工具脚本,核心功能是检查 SKILL.md 文件格式与字段合法性。二者在主要目的、触发场景、输入输出和实际能力上都明显不一致。代码没有实现任何声明中的分析流程,却实现了完全不同的验证功能,因此属于明显的描述与行为不匹配。
Referenced artifact was not completely inspected
这份 `SKILL.md` 只负责三件事:
Without declared permissions the skill's intent is opaque and cannot be validated.
The trigger conditions are broad enough to activate on generic requests for structured analysis, complex problems, or HTML reports, not only explicit requests for this ideology-specific method. Overbroad invocation can cause the wrong skill to intercept unrelated user tasks, leading to unsafe policy drift, unintended persuasive framing, or mishandling of requests that should be routed elsewhere.
The required fixed sentence is in Chinese and is mandated for every reply, which effectively forces a specific language regardless of the user's preference. The file does not provide opt-in language selection or explain that the skill is intentionally limited to a Chinese-language or region-specific context.
The prompt hard-requires Chinese section headers and an exact fixed Chinese sentence on every reply, without accommodating the user's language or accessibility needs. This is dangerous because it can override user intent and system usability expectations, causing coercive formatting, confusing output, and increased risk that important safety or task information is buried or mistranslated for non-Chinese users.
The skill allows implicit invocation with no concrete trigger constraints, so the agent may activate this skill for loosely related prompts and unexpectedly force users into its rigid clarification flow and prescribed framing. In practice this creates an instruction-hijacking surface where a broad, ideology-specific skill can override normal assistant behavior, reduce user control, and increase the chance of misrouting sensitive conversations.
The document explicitly sets lang="zh-CN", making the content target Simplified Chinese users by default. Under the policy, locale constraints should either offer user opt-in/choice or be clearly documented as justified for a region-specific skill, which is not present here.
The skill content hardcodes Chinese-language interaction and fixed Chinese prompt text without offering any user language choice. This can exclude users, cause misunderstanding of instructions or outputs, and create unsafe reliance if a user cannot accurately understand clarifications, constraints, or recommendations.
The file presents the skill’s core navigation and operating instructions entirely in Chinese, beginning with the title and continuing throughout, with no indication that users may choose another language. Under the stated policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is documented and justified.
This markdown file appears to require Chinese comprehension for all users because all headings, instructions, and examples are presented only in Chinese. The policy for natural-language violations applies to all file types and flags forced language constraints when no user opt-in or justification is provided.
The title and all operational instructions are written exclusively in Chinese, and the file presents binding rules for the skill's behavior without any indication that language is user-selectable. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless clearly documented as region-specific and justified, which is not stated here.
The file title and all prescribed response formats are written as mandatory Chinese-language instructions, and the examples require the assistant to reply using Chinese wording. There is no indication that language choice is optional or that the skill is intentionally limited to a Chinese-only, region-specific context.
This markdown file contains user-facing operational instructions exclusively in Chinese, and nowhere indicates that the skill is optional for Chinese-speaking contexts only or that users may choose another language. The policy explicitly calls for flagging language or locale constraints that are forced without user opt-in.
The file is entirely written in Chinese and operationally assumes Chinese as the interaction language, without stating that output language should follow user preference or offering a fallback. In a general-purpose skill, this can override user intent, reduce transparency, and cause users to misunderstand clarifications or analyses, which is especially risky for complex decision-support workflows.
The file is entirely in Chinese and does not provide a language choice, fallback, or explicit locale constraint. In a multi-user or multilingual deployment, this can cause users or downstream reviewers to misunderstand clarification prompts, consent boundaries, or analytical framing, leading to incorrect decisions or unsafe handling of sensitive situations.
The title and the entire instruction set are written as mandatory response rules in Chinese, including fixed output section labels and phrasing. This imposes a specific language/locale on all interactions without any opt-in or documented regional justification, which matches the natural-language policy violation criteria.
These lines require that if the report body is in Chinese, all user-visible template text must also be converted to Chinese and explicitly prohibit leaving English labels in place. This imposes a language constraint in the skill instructions without offering the user a language choice or opt-in, which matches the locale/language policy violation criteria.
L474-L476 明确要求当正文是中文时,封面标识、模块名、表头、按钮和通用字段名默认全部使用中文,并禁止保留英文模板文案。这属于自然语言层面的语言/locale 强制要求,但文档没有说明用户可选择其他语言,也没有给出仅适用于中文区域场景的限制。
The document explicitly sets lang="zh-CN", and the entire template content and instructions are written only in Chinese. For a general-purpose report template, this imposes a specific language/locale without opt-in or any stated regional justification, which matches the language/locale policy-violation category.
SQP-3 applies to all file types and covers language or locale policy violations. This markdown skill content appears to require Chinese comprehension throughout, but it does not provide user opt-in, alternative language availability, or a documented region-specific justification for the language constraint.
This markdown file contains user-facing instructional content exclusively in Chinese, and there is no indication that the skill is region-specific or that users can opt into this language. Under the policy rule, forcing a specific language without user choice is a natural-language policy violation.
This markdown file presents all instructions, triggers, and operational guidance exclusively in Chinese. Under the stated policy, forcing a specific language without user opt-in or a clearly justified regional scope is a natural-language policy violation.
This markdown file’s headings, instructions, and operational guidance are entirely in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking context. Under the policy rule for language/locale constraints, forcing a specific language without opt-in is a natural-language policy concern.
This file presents all operational guidance in Chinese and does not provide any user opt-in, alternative language path, or explanation that the skill is intended only for a Chinese-language audience. Under the policy, forcing a specific language without user choice can be a natural-language policy violation.