Back to skill

Security audit

视频分析处理

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its video-analysis purpose, but its arbitrary URL download and external upload flow can expose private network data or consume disk without enough safeguards.

Review carefully before installing. Use it only with videos or URLs you trust, avoid private/internal URLs, and assume selected local videos are uploaded to the provider for processing. Be aware that the client does not enforce the provider's retention promises and does not cap remote download size.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/utils/validator.js:3
Finding

Private-Network SSRF with External Data Upload

Content
View full analysis
{ if (this.__isRequireRedirect(response)) { redirectCount++; if (redirectCount > this.__opts.maxRedirects) { const err = new Error("Too many redirects"); this.__setState(this.__states.FAILED); this.emit("error", err); return reject(err); } const redirectedURL = /^https?:\/\//.test(response.headers.location) ? response.headers.location : new URL(response.headers.location, url).href; this.emit("redirected", redirectedURL, url); return getRequest(redirectedURL, getReqOptions(redirectedURL)); ``` `scripts/video2text/index.js:140-153`: ```js const presignedUrl = await video.getPresignedUrl(tokenValue, file); if (!presignedUrl || !presignedUrl?.url || presignedUrl.url === "") { throw new Error("获取预签名URL失败,请反馈给开发者"); } utils.printInfo("上传文件到安全空间..."); await upload.uploadFileT ...[truncated 2891 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/utils/helper.js:27
Finding

Unrestricted Remote Download Can Exhaust Local Disk Space

Content
View full analysis
{ const options = { retry: { maxRetries: constants.CREATE_MAX_ATTEMPTS, delay: constants.RETRY_INTERVAL, }, override: { skip: true, skipSmaller: true }, }; let progressLog = ""; const dl = new Downloader(url, path, options); ``` `scripts/utils/download.js:345-346`: ```js resolve({ name: this.__getFileNameFromHeaders(response.headers, response), total: parseInt(response.headers["content-length"]) || null, }); ``` `scripts/utils/download.js:547-549`: ```js this.__fileStream = fs.createWriteStream(this.__filePath, {}); } else { this.__fileStream = fs.createWriteStream(this.__filePath, { flags: "a" }); ``` ### Technical Analysis The downloader reads the remote `Content-Length` header but does not enforce a maximum permitted file size. It also accepts responses without `Content-Length`, allowing a server to stream data for an arbitrary duration. Downloaded bytes are written directly to a file without a cumulative byte limit. The configured timeout does not provide an effective total-download deadline when data continues to arrive, and retry behavior may repeat failed transfers. Cleanup does not adequately mitigate this issue because it only removes files older than 24 hours. A single active download can consume available storage before cleanup is relevant. Partial files may also remain after some failure conditions. ### Attack Path 1. An attacker supplies a URL controlled by the attacker as the video source. 2. The server returns an extremely large response, falsely reports a small `Content-Length ...[truncated 1181 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • 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
Findings (23)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明描述的是一个面向视频内容分析的技能,核心能力应包括获取视频、处理音频/语音、理解内容并生成总结或改写结果。但提供的代码片段仅实现了通用命令行参数解析:识别 flags、读取参数值、校验类型、处理默认值/重复参数/未知参数,并生成帮助文本。这与声明的主要用途完全不同。该代码既没有任何视频平台处理逻辑,也没有下载、ASR、NLP、翻译或内容理解相关行为。因此这是明显的描述与行为不一致。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个完整的视频分析技能,核心能力应包括获取视频、提取音频/文本并进行理解与生成。而提供的代码片段仅是 scripts/utils/download.js,一个底层下载工具类 Downloader。它通过 Node.js 的 http/https、fs 等模块实现文件下载到本地,支持 HEAD 获取大小、重定向处理、暂停/恢复、断点续传、重试、进度事件、停止后删除文件等。虽然“自动下载视频”可被视为声明中的一个支撑环节,但该代码没有任何与视频分析本身直接相关的实现,因此就该代码片段而言,实际行为与声明的主要目的存在显著不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description presents a full-featured video analysis/transcription skill. However, this code chunk is only a small utility that reads package metadata from the local filesystem to return the skill/package name. That behavior is materially different from the declared primary purpose and does not implement any of the advertised video-processing capabilities. While utility files can be supporting details, this specific chunk has no visible connection to the declared functionality beyond generic skill metadata access, so based on the supplied code alone, the description is not accurately represented by the actual behavior.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

声明描述的是一个完整的视频分析技能,但提供的代码片段只包含 token 校验辅助函数。虽然 token 校验可以是某些技能的配套实现细节,但该片段本身没有体现声明中的核心能力,且加入了明显与视频分析无关的营销式输出(官网与微信联系方式)。因此,这段代码与声明用途存在明显不一致,属于材料性不匹配。

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose describes a feature-rich video analysis skill, but the supplied code chunk is only a small utility module for validating URLs and local file paths. This is materially different from the declared primary purpose if assessed on this chunk alone. While such validation could be a supporting detail in a larger system, the local file path checking via fs is an implementation capability not reflected in the description, and none of the advertised analysis behaviors appear here.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README defines very broad natural-language triggers like '总结一下这个视频' and automatic conversion rules that map ordinary user phrasing directly into tool execution. In an agent setting, this can cause over-triggering or unintended invocation when a user is merely discussing a video rather than explicitly consenting to upload/process it, which is especially sensitive because the skill sends content to a remote service and may incur billing.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · README.md (reported line 99)May include surrounding context.

md
4. 同时传入文件路径与任务ID,优先执行 `--id`,忽略 `--file`
5. 无自定义 prompt 时,默认完整转录视频全部文字

| 用户自然语言指令                                         | 自动生成命令                                                                                                |
| -------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| 视频提取 https://example.com/video.mp4 中的文字          | `node scripts/video2text/index.js --file "https://example.com/video.mp4"`                                   |
| 把本地 /path/to/your/video.mp4 改成小红书风格的文案      | `node scripts/video2text/index.js --file "/path/to/your/video.mp4" --prompt "改写成小红书风格的文案"`       |

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is designed to trigger on very broad natural-language requests around video analysis and content generation, which increases the chance of over-invocation on loosely related prompts. In context, that can cause unintended upload of user-provided local files or remote URLs to a third-party service, creating avoidable data exposure and billing risk.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The examples and mapping rules encourage open-ended natural-language matching and automatic command construction, including reuse of --id last. In a skill that uploads media and can reuse prior task context, broad matching can trigger actions on the wrong asset or send data externally without sufficiently specific user confirmation.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 125)May include surrounding context.

md
当用户用自然语言下达指令时,按以下映射生成命令,保证识别与执行一致:

| 用户自然语言指令                                     | 生成的命令                                                                                                  |
| ---------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
| 提取 https://example.com/video.mp4 里的文字          | `node scripts/video2text/index.js --file "https://example.com/video.mp4"`                                   |
| 总结这个视频的核心观点 https://example.com/video.mp4 | `node scripts/video2text/index.js --file "https://example.com/video.mp4" --prompt "总结这个视频的核心观点"` |

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

The description forces a specific language context in the skill metadata, which may violate language/locale policy when no opt-in or alternative is provided. Because this is user-facing descriptive text rather than a clearly justified region-specific compliance constraint, it presents a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This JavaScript file contains natural-language comments and multiple user-facing error/help strings in Chinese, indicating the skill forces a specific language experience. Under the policy, language constraints should be opt-in or clearly justified as region-specific, which is not evident 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 parser throws errors such as "参数 ... 缺少值" and generates help text like "用法" and "选项" entirely in Chinese. Because the file provides no mechanism for locale selection and no justification for a Chinese-only audience, this 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
97% confidence
Finding

This code contains natural-language content that forces a specific language for documentation and runtime status/error messages, such as the Chinese comment and download progress messages. The file does not offer user opt-in, localization, or any documented justification that the skill is intentionally limited to a Chinese-speaking context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This code performs an outbound HTTPS POST and includes both serialized request data and a TOKEN header, which can transmit user or system data to a remote service. In this file there is no confirmation prompt, user-facing warning, or explanatory comment/docstring disclosing that network transmission and token use will occur.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The warning and promotional strings are hard-coded entirely in Chinese, which imposes a specific language on all users. There is no indication of locale selection, fallback behavior, or user opt-in to Chinese, so this violates the language/locale policy criterion.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The comment makes a privacy and data-handling guarantee ('not leaked, not retained, automatically deleted after processing') that is not enforced anywhere in this code. In a skill that uploads user videos for transcription and analysis, this mismatch can lead to sensitive media being retained in object storage indefinitely or handled contrary to user expectations, creating privacy, compliance, and trust risks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This code exposes user-facing help text, flag descriptions, and examples exclusively in Chinese. That imposes a specific language on users without offering a locale choice or documenting that the tool is intentionally region-specific, which matches the policy-violation criteria for language/locale constraints.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The feature list includes '全文中英互译', and later examples explicitly suggest '全文翻译成英文' as a default-type instruction. This hard-codes a specific language target in the skill description without indicating that users can choose their preferred locale or language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JavaScript file contains natural-language comments and user-visible log/error strings exclusively in Chinese, including operational disclosures about deleting expired files. The policy specifically flags language or locale constraints when a skill forces a specific language without user opt-in, and there is no indication here of any language selection or justified locale limitation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

The user-visible error strings in this file are written only in Chinese, including operational and credential-related failures. This hard-codes a specific language/locale without any opt-in or documented justification, which is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The banner string is hard-coded in Chinese, which imposes a specific language choice in the user-facing output. The file does not indicate any user opt-in, language selection mechanism, or documented region-specific requirement that would justify this locale constraint.

Content

No source excerpt is available for this finding.