Back to skill

Security audit

Openai Whisper

Security checks for vulnerabilities and agentic risk

Overview

The skill appears intended for local Whisper transcription, but its instructions conflict about API keys and API use while granting command and file-write capability.

Review before installing. Use it only as a local Whisper CLI transcription helper, avoid providing any API key unless the publisher clearly documents a real remote mode, and choose explicit safe input and output paths because transcripts may be written to disk.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (4)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill documentation is internally inconsistent about whether an API key is required and whether operation is purely local. That can mislead users into supplying sensitive credentials unnecessarily or trusting a local-only privacy posture when the skill may involve external API usage, which creates avoidable secret exposure and data-handling risk.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The invocation guidance is overly broad and appears copied from a generic API-integration skill, making it unclear when this skill should be used and what actions it may perform. Ambiguous scope increases the chance that an agent invokes tools such as exec, read, or write in inappropriate contexts, which can lead to unintended command execution or mishandling of user data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The documented command writes an output file but does not clearly warn the user about filesystem side effects, output location, or overwrite behavior. In an agent context with write capability, missing side-effect disclosure can cause unexpected file creation, accidental overwrites, or exposure of transcription output on disk.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

The documentation gives conflicting signals about network behavior, which undermines a user's ability to assess privacy and exposure. For a speech-to-text skill handling potentially sensitive audio, ambiguity around network use can cause users to process confidential data under false assumptions.

Content

No source excerpt is available for this finding.