Back to skill

Security audit

Qwen Video Analyzer Free

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed Qwen/DashScope video-analysis helper, but users should treat videos, URLs, and the API key as sensitive because processing uses an external service.

Install only if you are comfortable sending the selected video content or public video URL to DashScope/Qwen for analysis. Do not use internal, authenticated, pre-signed, or confidential URLs, protect the DashScope API key file, and prefer a virtual environment with a pinned dashscope version.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (1)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:226
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, line 226
Vulnerability Type: Unpinned and mutable third-party dependency
Risk Level: Medium

Vulnerable Code Snippet:

markdown
| dashscope | Python SDK | Required | pip install dashscope |

Technical Analysis

The installation instruction uses pip install dashscope without specifying a reviewed version or an integrity hash. Consequently, the installed artifact depends on the package version currently resolved by pip and on the configured package index.

This does not establish that the named dependency is malicious. However, it creates a supply-chain exposure because future releases are not covered by the original Skill review. A compromised package release, package repository, dependency account, or local pip index configuration could cause an unexpected artifact to be installed. Transitive dependencies are also left unlocked and unverified.

Attack Path

  1. An attacker compromises a package release, an upstream dependency, a configured package index, or the dependency publication process.
  2. The attacker publishes a malicious or otherwise compromised version that satisfies the unrestricted installation command.
  3. A user follows the Skill documentation and executes pip install dashscope.
  4. Pip resolves and installs the compromised version because no version constraint or integrity hash is enforced.
  5. Malicious code may execute during installation or when the package is subsequently imported by the video-analysis workflow.

Impact Assessment

Successful exploitation could execute attacker-controlled code with the privileges of the user running pip or the analysis process. Depending on those privileges and the environment, the attacker could access user-readable files, application credentials such as the configured DashScope API key, video inputs, network resources, and writable project or user directories.

The issue does not indep ...[truncated 171 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin dashscope to a specifically reviewed version rather than installing an unrestricted latest release.
  2. Place direct and transitive dependencies in a lock file generated from a trusted environment.
  3. Require package integrity hashes, such as with a hash-locked requirements file and pip install --require-hashes.
  4. Configure pip to use an explicitly trusted package index and prevent unintended fallback to untrusted or privately configured repositories.
  5. Review dependency updates before changing the pinned version, including release provenance, dependency changes, and known vulnerabilities.
  6. Install and run the dependency in an isolated virtual environment or container under a non-privileged account.
  7. Restrict the runtime environment so it can access only the required videos, configuration fields, and network endpoints.
  8. Document a reproducible installation command, for example:
bash
python -m pip install --require-hashes -r requirements.txt
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
79% confidence
Finding

该技能的名称、描述、示例提示词及整体文档均固定为中文表述,且未说明是否支持其他语言或允许用户自行选择语言输出。对于面向“任意 AI Agent”的通用技能,这种未声明的单一语言约束可能构成语言/locale 策略问题。

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill advertises analysis of remote video URLs but does not clearly disclose that the provided URL and associated video content may be sent to an external DashScope/Qwen service for processing. This creates a privacy and data-handling risk because users or calling agents may supply sensitive internal, pre-signed, or otherwise confidential URLs without informed consent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Lines L041-L065 describe generic operation modes such as 创建/查询/修改/删除, 修改/重置/导入, and 导出/保存/转换, which contradict the rest of the file's concrete behavior: analyzing local or remote videos and printing results to the terminal. This is not merely incomplete documentation; it actively describes broader state-changing capabilities that the skill's own usage examples and purpose do not support.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger condition is overly broad, effectively saying to use this skill whenever AI model calling, agent orchestration, or LLM applications are involved. In an agent ecosystem this can cause the skill to activate for unrelated tasks, increasing the chance of unnecessary exec usage, unintended remote URL handling, or accidental transmission of user-provided content to external services.

Content

No source excerpt is available for this finding.