Security audit

transcribe-video

Security checks across malware telemetry and agentic risk

Overview

This video transcription skill is mostly purpose-aligned, but it deserves review because it can upload local audio to an external API and its helper script may overwrite or delete a same-named WAV file next to the video.

Install only if you are comfortable sending videos without embedded subtitles to the configured OpenAI or Azure-compatible transcription provider. Protect ~/.transcribe_video.env with restrictive permissions, verify OPENAI_API_BASE points to a trusted endpoint, and check for an existing same-named .wav file beside the video before running because the helper may upload and delete it.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (7)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 93% confidence
Finding: The skill invokes shell commands, reads environment-based credentials, and writes files, but does not declare those capabilities. Hidden capabilities reduce transparency and prevent informed consent or policy enforcement by the host system. In this context, the omission is risky because the skill is presented as a simple local transcription utility while it can access secrets and modify the filesystem.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 96% confidence
Finding: The documented behavior goes beyond local transcription by uploading extracted audio to an external API, reading credentials from the home directory, and writing output files locally. This mismatch can mislead users into exposing sensitive audio under the assumption that processing is purely local. Because video content may contain confidential conversations, the undisclosed external transfer materially increases privacy and compliance risk.

Description-Behavior Mismatch

Medium

Confidence: 90% confidence
Finding: The manifest frames the skill as handling local video files, but the implementation sends audio to a remote transcription service when subtitles are absent. 'Local file' wording describes the source of the input, not the destination of the data, so users may reasonably infer no off-device transfer occurs. That ambiguity is dangerous for sensitive recordings and regulated data.

Context-Inappropriate Capability

Medium

Confidence: 88% confidence
Finding: The skill instructs access to an API key stored in the user's home directory for a task described as ordinary local transcription. Accessing persistent credentials expands the trust boundary and can normalize broad secret access for a routine workflow. In a skill ecosystem, this increases the chance of secret exposure or misuse if the skill is modified, compromised, or misunderstood.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The skill is presented as operating on local video files, but the implementation sends extracted audio to OpenAI or Azure transcription services. This mismatch can cause users to disclose sensitive local media under the false impression that processing is purely local, creating privacy and compliance risk.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The description omits two material side effects: sending audio to an external service and writing transcript files next to the source video. Without that warning, users cannot make an informed decision about privacy, storage location, or data retention. This is especially risky for videos containing personal, proprietary, or regulated speech.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The code transmits audio derived from a local video to a third-party transcription API without any explicit warning, consent step, or privacy notice. In a transcription skill, the content may include confidential conversations, regulated data, or personal information, so silent network transfer materially increases privacy risk.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.