whitehole-audio
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- EXTERNAL_DOWNLOADS (HIGH): The skill directs the cloning of a repository from an untrusted GitHub account (
bmorphism/WhiteHole). This source is not listed in the [TRUST-SCOPE-RULE] whitelist. - COMMAND_EXECUTION (HIGH): The instructions include
sudo cp -Randsudo launchctl kickstart, which require root privileges. Granting an agent or skill the ability to execute commands with elevated privileges on sensitive system directories (/Library/Audio/Plug-Ins/HAL/) is a high-risk operation. - REMOTE_CODE_EXECUTION (HIGH): The sequence of
git clonefollowed byxcodebuildand thensudoinstallation constitutes a Remote Code Execution (RCE) vector, as the contents of the untrusted repository are compiled and executed with system-level permissions. - Persistence Mechanisms (HIGH): Installing a driver into
/Library/Audio/Plug-Ins/HAL/ensures that the code remains active across system reboots, serving as a persistence mechanism. - Metadata Poisoning (MEDIUM): The skill uses dense, pseudo-scientific terminology (e.g., "GF(3) Balanced Triad", "Cat# = Comod(P)", "bicomodule") which may serve to obscure the simple functional nature of the script or mislead users/analyzers regarding its true purpose.
Recommendations
- AI detected serious security threats
Audit Metadata