Back to skill

Security audit

children-science-video

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a disclosed Chinese children’s science video workflow, but its trigger is too broad and can run a full file- and image-generation workflow when a user merely pastes related text.

Install only if you want a Chinese-language children’s science short-video asset workflow. Treat pasted science or safety text carefully, because the skill says to auto-run on broad inputs; confirm before letting it create files or use image-generation credits.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (14)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个面向内容生产的端到端自动化技能,核心能力应是根据输入文案生成多种短视频产物。而实际代码没有任何生成逻辑、模型调用、图片处理、封面生成、社媒文案输出或自动触发机制。它只是一个辅助质检脚本,针对现成 Markdown 分镜表中的“配音文案”列做字数规则校验。虽然这可被视为分镜流程中的支持性实现细节,但如果把该代码块作为技能行为代表,其主要目的与声明的主要目的明显不一致,因此应判定为描述与行为不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个内容生成型多步骤短视频制作技能,核心能力应包括根据用户提供的科普文本自动产出多种创作素材与图片。而实际代码是单一的 PDF 排版导出脚本,输入为 Markdown 文件路径,输出为 PDF 文件;它处理字体、样式、标题、列表、emoji 清理和页脚,并未见任何自然语言生成、分镜规划、提示词优化、封面设计或图像生成逻辑。虽然页脚和标题中出现“儿童科普短视频·双平台发布文案”等字样,说明它可能是整套流程中的一个交付环节,但就该代码块本身而言,其主要目的与声明的整体技能能力严重不符,因此应判定为描述与行为不匹配。

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The auto-trigger condition is extremely broad: any pasted science, safety, or knowledge-related text can activate the skill, even when the user did not ask for video generation. This creates prompt-scope hijacking risk, where unrelated conversations or sensitive pasted content are unexpectedly transformed, causing unintended processing, file generation, or downstream tool usage.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The rule 'any science copy triggers' lacks boundaries, exclusions, and confirmation gates, so the skill can seize control of prompts far outside the user's intended task. In an agent environment, this broad interception is dangerous because it can override normal routing and cause the system to process arbitrary educational or safety text with unnecessary file and tool operations.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
72% confidence
Finding

The skill instructs the agent to read multiple local reference, asset, and script files, but it does not explicitly declare a constrained tool scope such as allowed file paths or read-only permissions. In practice, this can lead to broader-than-expected file access if the runtime grants generic file_read capability, increasing the risk of unintended local data exposure or misuse.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

L23 规定“中文输出”,属于语言/locale 约束,但文件中未见用户可选择其他语言的机制,也未说明该限制基于特定地区合规或明确业务边界。根据规则,这类未经用户选择的强制语言要求属于自然语言政策问题。

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document is written entirely in Chinese and prescribes fixed Chinese descriptors such as '一个中国小男孩' and Chinese text elements, without indicating that language or locale is optional. This can violate language/locale policy when a skill implicitly requires a specific language without user opt-in or justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This markdown file contains prescriptive guidance entirely in Chinese, including imperative requirements such as "必须给出", but nowhere indicates that Chinese is optional or that the skill is intended only for a Chinese-language workflow. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This markdown file contains prescriptive content entirely in Chinese and specifies Chinese platform copy conventions, Chinese fonts, and output structure, but does not indicate that language selection is optional or user-driven. Under the policy rule, forcing a specific language or locale without opt-in can be a natural-language policy violation unless clearly documented as a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The guidance requires Chinese-character counts and provides fixed Chinese phrasing examples for voiceover structure, which effectively constrains the skill to Chinese output. The file does not indicate that language selection is optional or that the Chinese-only requirement is justified by a region-specific use case.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file contains user-facing natural-language strings that assume Chinese as the only language, including the module description, usage context, and validation terminology. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The script prints all errors, status labels, and result summaries in Chinese, which enforces a single language experience for all users. Because the file does not offer language selection or clearly document a justified region-specific constraint, this violates the natural-language policy requirement.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes an automatically triggered end-to-end children's science video production skill that creates video copy, storyboard tables, prompts, covers, platform-ready post text, and batch image generation. This file's documented and actual behavior is much narrower: it only converts an existing Markdown document into a Chinese PDF for dual-platform publishing copy.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The footer and document metadata show this code is specialized for a 'dual-platform publishing copy' PDF, while the manifest emphasizes a broader package including Douyin/Xiaohongshu copy plus multiple non-PDF assets and batch 9:16 images. The implementation here supports only one narrow deliverable and encodes that narrower purpose directly into output content.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.