Security audit
Security checks for vulnerabilities and agentic risk
The skill says it uses a local Whisper server, but its script sends audio files to a remote HTTP server instead.
Review before installing or using this skill. It may expose any audio you transcribe to sphinx.espuny.net over unencrypted HTTP, despite presenting itself as a local server bridge. Use only non-sensitive test audio unless the author updates the script and documentation to clearly disclose the destination, use secure transport, and provide a genuine localhost-only configuration.
The skill advertises a local speech-to-text bridge, but the script uploads the provided audio file to a remote host on the public internet. This is a significant trust and data-handling mismatch: users may provide sensitive recordings under the assumption that processing stays local, when in fact the audio leaves the machine and is exposed to remote retention, interception, or misuse.
The header comment explicitly states that the bridge redirects Whisper CLI calls to a GPU server, which contradicts the metadata claiming a local Whisper-based service. This inconsistency is dangerous because it suggests the package is presented in a misleading way, increasing the chance that users unknowingly expose private audio to a remote system.
If audio content is sent over HTTP for transcription, failing to warn users is a meaningful security and privacy issue because speech may contain secrets, personal data, or regulated information. Even when the destination is localhost, users and integrators need explicit disclosure that audio leaves the calling process and is transmitted to another service endpoint.
The script transmits the user-supplied audio file to an external server without any user-facing warning, consent prompt, or disclosure. In the context of a speech-to-text skill, audio commonly contains sensitive personal, business, or authentication information, so silent exfiltration to a third party materially increases privacy and compliance risk.
The script uses plain HTTP to send audio to the remote transcription endpoint, so the recording and associated parameters can be intercepted or modified by anyone with network visibility. Because the payload is raw user audio, this creates a direct confidentiality and integrity risk in addition to the undisclosed remote transfer.
