Security audit

Byted Mediakit Audio

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent MediaKit audio-processing skill, with the main caution that cloud mode can send media URLs or local media files to a remote service.

Install only if you trust the MediaKit CLI and understand that cloud mode may upload local media files or submit media URLs to a remote service. Avoid using confidential, regulated, internal, or signed-link media unless you have authorization and are comfortable with third-party processing; prefer local mode where supported.

SkillSpector

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 (5)

Intent-Code Divergence

Medium

Confidence: 90% confidence
Finding: The documentation states the feature only supports public HTTP/HTTPS URLs, but elsewhere says it supports local mode and can be forced with --local. In a shell-permission skill, this ambiguity can cause unsafe assumptions about whether local files may be accessed, potentially expanding the data-access surface and leading users or agents to probe unintended local resources.

Vague Triggers

Medium

Confidence: 82% confidence
Finding: The skill description says it should trigger whenever the user needs MediaKit CLI audio capabilities, which is a broad activation rule without clear boundaries or disambiguation criteria. In an agent system with shell permission, overly broad triggering can cause the skill to activate in unintended contexts, increasing the chance of unnecessary command execution or misuse of audio-processing operations on untrusted inputs.

Natural-Language Policy Violations

Medium

Confidence: 74% confidence
Finding: The skill content is written entirely in Chinese and may effectively constrain interaction or documentation to a specific language without stating that this should follow the user's preference. This is primarily a safety and usability issue: language forcing can cause the agent to misunderstand user intent, omit important warnings or confirmations, and execute shell-capable actions without the user fully understanding what the skill does.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The skill is cloud-only and requires publicly accessible media URLs, which means user-provided audio or video is sent to an external service and may be exposed beyond the local environment. Although the parameter table mentions that URLs must be public, the description does not clearly warn about privacy, confidentiality, and third-party processing implications, so users may unintentionally submit sensitive media.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The documentation states that in cloud mode, local media paths are automatically uploaded and remote URLs are forwarded, but it does not clearly warn that user-provided local files and referenced media will leave the local environment and be sent to a cloud service. In a shell-capable agent skill, this increases the risk of unintended data exfiltration because an agent may pass sensitive local paths or internal URLs without explicit user awareness or consent.

Static analysis

No suspicious patterns detected.