Security audit

Azure语音转写免费版

Security checks across malware telemetry and agentic risk

Overview

The skill is mainly an Azure speech-to-text helper, but its activation instructions are too broad and partly conflict with that purpose while it uses command execution, Azure credentials, and cloud audio processing.

Install only if you intend to use it for Azure-based audio transcription, and avoid letting it handle unrelated translation or localization tasks. Treat audio recordings, Blob URLs, and Azure subscription keys as sensitive; confirm the target Azure resource, storage access, and output path before running generated Python or shell commands.

SkillSpector

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: The skill documentation broadens a narrowly defined transcription tool into generic create/query/modify/delete-style operations via abstract parameters such as input_params, config_options, and output_format. In an agent setting, this can cause the model to over-apply the skill to unrelated tasks and execute actions outside the user’s intended scope, increasing the chance of unsafe tool use.

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: The trigger conditions instruct use of this skill for text translation, multilingual conversion, and localization, which conflicts with the rest of the document describing speech transcription. This mismatch can cause an agent to invoke the skill in the wrong context, potentially sending unintended content or performing cloud processing the user did not request.

Vague Triggers

High

Confidence: 89% confidence
Finding: The top-level description says to use the skill whenever efficiency, automation, batch processing, or workflow optimization is needed, which is far broader than audio transcription. In an agent marketplace, overly broad triggers can hijack unrelated requests and lead to inappropriate tool invocation and data transfer to external services.

Vague Triggers

High

Confidence: 90% confidence
Finding: The keyword-based coverage claim includes vague fragments and broad scenario terms rather than precise trigger conditions. This increases the risk that an agent selects the skill based on loose keyword overlap rather than actual user intent, which is especially dangerous because the skill can execute code and contact external cloud services.

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The trigger condition both conflicts with the tool’s purpose and remains underspecified, encouraging activation for unrelated text-processing tasks. This can misroute user workflows into a transcription skill with exec capability, creating avoidable exposure of data and unnecessary command execution.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The skill does not clearly warn that audio URLs and their contents are submitted to Azure cloud services for processing. Users and agents may therefore disclose sensitive recordings or private Blob URLs without informed consent, which is a meaningful privacy and compliance risk in a cloud-processing workflow.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.