Back to skill

Security audit

fun-denoise

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent Alibaba Cloud audio denoising skill that uploads user-chosen audio for processing, with privacy and dependency cautions but no hidden persistence or unrelated access.

Install this in a virtual environment, consider pinning dependency versions, and only process audio you are authorized to upload to Alibaba Cloud/DashScope because recordings may contain sensitive personal, business, or biometric information.

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:39
Finding

Unpinned Third-Party Dependencies Permit Supply-Chain Drift

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:39
Vulnerability Type: Unpinned third-party dependencies
Risk Level: Medium

Vulnerable code:

bash
pip install dashscope websocket-client

Technical Analysis

The documented installation command installs dashscope and websocket-client without exact version constraints or package hashes. Consequently, installations performed at different times may retrieve different artifacts that were not part of this audit.

Package installation can execute package-controlled build or installation logic, while the installed libraries are subsequently imported and executed by the Skill. If an upstream account, release process, package index, or dependency is compromised, following this instruction could introduce attacker-controlled code. This finding concerns the absence of reproducible dependency controls; no evidence was found that the currently named packages are malicious.

Attack Path

  1. An attacker compromises an upstream package publisher, distribution account, release pipeline, package index, or transitive dependency.
  2. The attacker publishes a malicious release under one of the dependency names or alters an artifact distributed for an otherwise valid release.
  3. A user follows the documented unversioned pip install command.
  4. pip resolves and downloads the attacker-controlled artifact because no reviewed version or integrity hash is required.
  5. Malicious code executes during package installation or when denoise_cli.py imports and uses the dependency.

Impact Assessment

Successful exploitation could execute arbitrary code with the privileges of the user running pip or the Skill. Depending on those privileges, the attacker could read or modify accessible files, access environment variables such as DASHSCOPE_API_KEY, intercept uploaded audio, alter denoising results, or initiate network connections. The scope is limited by the operat ...[truncated 90 chars]

Remediation
View remediation

Remediation Suggestions

  1. Create a dependency manifest containing exact, reviewed versions of dashscope, websocket-client, and all transitive dependencies.
  2. Generate a lock file through a reproducible dependency-management workflow and commit it to the project.
  3. Record cryptographic hashes for every permitted distribution and install with hash verification, for example:
    bash
    python -m pip install --require-hashes -r requirements.txt
    
  4. Configure an explicit trusted package index or an internally controlled artifact repository rather than relying on implicit index configuration.
  5. Run dependency vulnerability and provenance checks in CI, review updates before changing the lock file, and use automated update tooling with mandatory approval.
  6. Install and run the Skill in an isolated virtual environment or container under a non-privileged account, exposing only the files and credentials required for audio processing.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (8)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill clearly instructs users to send audio to Alibaba Cloud/DashScope over a WebSocket API, but it does not prominently warn that uploaded recordings leave the local environment and are processed by a remote third-party service. This creates a privacy and data-handling transparency gap that can lead users to submit sensitive conversations, meetings, or personal recordings without informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The code reads an API key from configuration and injects it into the Authorization header for outbound requests. Although this is expected for authenticated SDK behavior, the file does not provide a user-facing disclosure that credentials are being used and sent to a remote service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

This code establishes a remote WebSocket connection and later sends user-provided audio frames for processing, which is a network operation involving potentially sensitive user data. While the class docstrings describe functionality, there is no confirmation prompt or explicit user-facing disclosure in the code warning that audio content will be sent to an external service.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The CLI imports dashscope and a Denoise client implementation, indicating processing is performed by a remote service rather than locally on-device. For an audio-processing skill, this materially changes the trust boundary because user audio may leave the host and be handled by a third party despite the description implying a simple denoise utility.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code reads arbitrary audio from disk and sends it frame-by-frame to an external denoising service without any explicit warning, consent gate, or privacy notice at the point of transfer. Audio often contains sensitive personal, business, or biometric information, so silent remote transmission increases the risk of unintended data disclosure and noncompliant handling of user content.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The natural-language description forces a specific language presentation for users, and there is no indication that the skill is region-specific or that users can opt into another language. Under the stated policy, language constraints should either be optional or clearly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
81% confidence
Finding

User-facing descriptions, logs, and help text are largely in Chinese, which effectively forces a language/locale choice on users without opt-in. The file does not state that it is intended only for a Chinese-speaking or region-specific audience.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

The manifest frames the skill as an audio denoising service for removing background noise, but the CLI explicitly supports a '--no-denoise' mode described as '仅转换格式' (format conversion only). That adds a standalone audio conversion capability not justified by the denoising-focused description.

Content

No source excerpt is available for this finding.