Back to skill

Security audit

Fun-ASR-File

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent cloud transcription helper, but users should only run it on audio they are willing to send to Alibaba Cloud DashScope.

Install in a controlled Python environment, pin or review the DashScope SDK version, and use the skill only for audio files you are allowed to send to Alibaba Cloud DashScope. Ask for explicit confirmation before transcribing sensitive recordings.

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:32
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:32
Vulnerability Type: Unpinned package installation from a mutable package index
Risk Level: Medium

bash
pip install dashscope

Technical Analysis

The installation instructions retrieve dashscope without a version constraint, lockfile, package hash, or explicit trusted repository. Consequently, the package resolved during installation can change after the skill has been reviewed.

This creates supply-chain exposure if a future package release, package-publishing account, configured Python package index, or upstream distribution channel is compromised. Python packages can execute code during installation and whenever imported by scripts/cli.py, so a malicious resolved package could run with the privileges of the user operating the skill.

No evidence was found that the current dashscope package is malicious. The vulnerability is the absence of controls that ensure users install the same reviewed dependency artifact.

Attack Path

  1. An attacker compromises the package publisher, publishing credentials, upstream package index, or a package source configured in the victim's pip environment.
  2. The attacker makes a malicious dashscope release or replacement artifact available for resolution.
  3. A user follows the documented pip install dashscope command.
  4. Because no version or hash is enforced, pip resolves and installs the attacker-controlled artifact.
  5. Malicious code executes during installation or when scripts/cli.py imports dashscope.
  6. The code operates with the invoking user's permissions and may access the DASHSCOPE_API_KEY, local audio files, and other data available to that account.

Impact Assessment

Successful exploitation could provide arbitrary code execution under the account performing installation or transcription. The reachable scope includes environment variables, submitted audio files, files readable by t ...[truncated 219 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin dashscope to a reviewed, exact version rather than resolving the latest available release.
  2. Maintain dependencies in a lockfile or requirements file containing cryptographic hashes.
  3. Install with hash verification, for example through pip install --require-hashes -r requirements.txt.
  4. Document and enforce the official trusted package index instead of inheriting arbitrary pip index configuration.
  5. Review dependency updates before changing the pinned version and use automated vulnerability and provenance checks.
  6. Run installation and transcription as an unprivileged user in an isolated environment with access limited to the required audio file and API credential.
Vulnerability Patterns
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Tp4

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

代码的核心目的与描述大体一致:它确实是一个针对本地音频文件的非流式语音识别脚本,调用阿里云 DashScope/FunASR 服务完成转写。但描述中的两个具体能力与代码不符。首先,代码将识别参数固定为 format='wav' 和 sample_rate=16000,没有检测输入格式、转码或采样率转换逻辑,因此“支持自动格式转换”不成立。其次,描述称“适合批量文件转写场景”,但当前实现只接收并处理单个文件路径,没有循环、目录处理或批量任务支持。除这些差异外,代码没有表现出额外未声明的敏感能力。

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
81% confidence
Finding

The skill requires an environment secret (DASHSCOPE_API_KEY) and appears to rely on environment access, but it does not declare any tool scope or permissions boundary. This creates an authorization and transparency gap: platforms or reviewers cannot clearly tell that the skill needs secret-bearing environment access, which increases the chance of over-privileged deployment and unsafe secret exposure.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill handles local audio files but does not clearly disclose that those files will be transmitted to Alibaba Cloud DashScope for transcription. This is dangerous because users may assume processing is local and unknowingly send sensitive conversations, personal data, or regulated information to a third-party cloud service.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The activation condition triggers on any user-sent audio file, even when the user did not explicitly request transcription. In context, this is more dangerous because the skill sends content to an external API, so an overly broad trigger can cause unintended exfiltration of private audio merely from file sharing.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script sends a user-supplied local audio file to Alibaba Cloud DashScope for transcription, but it does not clearly disclose that the file contents leave the local machine and are transmitted to a third-party service. This can expose sensitive voice data, background conversations, or embedded personal information without informed user consent, which is especially risky in a local-file batch transcription workflow where users may assume processing is local.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The user-facing prompt and error text are hard-coded in Chinese (请输入音频文件路径, 文件不存在) with no option to select another language. This creates a natural-language policy concern because the skill imposes a locale on all users without opt-in or justification.

Content

No source excerpt is available for this finding.