Security audit

transcription

Security checks across malware telemetry and agentic risk

Overview

This transcription skill does what it claims, but it uploads potentially sensitive media to a hardcoded plain-HTTP local-network endpoint without enough user control or privacy disclosure.

Install only if you control and trust the service at `192.168.0.11:8080` and accept that recordings will be sent there over plain HTTP. Avoid sensitive, confidential, medical, legal, or business recordings unless the endpoint is secured and its retention/logging behavior is known. A safer version would use a configurable HTTPS endpoint and require explicit confirmation before uploading media.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (8)

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The phrase 'Send me an audio or video file and I'll transcribe it automatically' is broad enough to trigger on ordinary media-sharing requests without clearly signaling that a skill will activate and potentially upload the file elsewhere. In a transcription skill, broad invocation increases the risk of inadvertent processing of private recordings, especially when combined with undisclosed network transmission.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The skill encourages users to upload audio/video but does not warn that the media will be transmitted to a Whisper API endpoint, which is a significant privacy and data-handling omission. Because audio and video often contain sensitive personal, business, or biometric information, failing to disclose network transfer materially increases the risk of non-consensual exposure.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The guide explicitly states that user audio/video files are sent to a locally hosted Whisper API endpoint, but it does not disclose data handling, retention, access controls, or privacy implications. Even on a private network, transmitting potentially sensitive recordings to another service can expose confidential content if the endpoint is misconfigured, logged, shared, or reachable by unintended parties.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The script uploads the provided media file to a remote service over plain HTTP, which exposes the full audio or video contents to interception or modification by anyone with network visibility. Because transcription inputs commonly contain sensitive voice data, meetings, or personal recordings, sending them without transport encryption or an explicit user warning creates a real confidentiality and integrity risk.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The function uploads user-provided audio/video-derived content to a remote HTTP Whisper API endpoint without any explicit disclosure, consent flow, or transport security. This can expose sensitive speech content and metadata to unintended parties, especially because the configured endpoint uses plain HTTP to a network host.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: This alternate transcription path also transmits audio content to the same remote Whisper service without disclosure or consent controls. Because users may invoke the simpler CLI path expecting local processing, the hidden exfiltration risk is significant for private recordings.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The timestamped transcription path sends the full media content to a remote endpoint without warning and may additionally return richer metadata such as word timing. For sensitive recordings, this increases privacy exposure while still using an insecure HTTP endpoint.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The script uploads the full contents of the user-supplied audio file to a remote HTTP transcription service at a hardcoded private-network endpoint without any disclosure, consent prompt, or indication that data leaves the local machine. This is risky because audio/video recordings often contain sensitive personal, business, or regulated information, and the use of plain HTTP also removes transport confidentiality, increasing exposure on local networks.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.