ClawHub
Back to skill

Security audit

llm-provider Whisper

Security checks for vulnerabilities and agentic risk

Overview

The skill claims to be a local Whisper transcription tool, but its instructions are broad and inconsistent about API keys, external services, and command/file access.

Review before installing. Use it only if you are comfortable granting command execution and file read/write access, and clarify whether it is meant to run local Whisper only or contact external APIs before providing audio files or credentials.

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 (4)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The skill documentation makes conflicting claims about whether an API key is required, alternating between 'no API key' and mandatory API-key setup. This can mislead users and downstream agents about whether data will stay local or be sent to an external provider, causing unintended credential use, privacy exposure, and unsafe execution assumptions.

Intent-Code Divergence

Low
Confidence
83% confidence
Finding
The FAQ says no extra software is needed even though the skill is presented as a Whisper CLI and command-execution based tool, which implies local binaries or dependencies. This can cause users or agents to invoke nonexistent or substitute tools unexpectedly, increasing the chance of unsafe fallback behavior, broken workflows, or accidental execution in the wrong environment.

Vague Triggers

High
Confidence
89% confidence
Finding
The activation guidance is overly broad, telling agents to use the skill for generic API integration, webhooks, and system connection tasks that are unrelated to speech-to-text. In an agent environment, this can cause the skill to trigger in inappropriate contexts and gain access to exec/write capabilities for tasks outside its intended scope, expanding attack surface and enabling unintended system-impacting actions.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The skill advertises file writing, external API/service use, and command execution without a prominent upfront warning about side effects, data exposure, or system integrity risks. In agentic workflows, missing disclosure reduces informed consent and can lead users to approve actions that modify files, execute commands, or transmit data without understanding the consequences.

Static analysis

No suspicious patterns detected.