branded-ad-edit

The code is primarily a legitimate audio-generation and normalization utility. It intentionally sends the configured prompt and locally sourced ElevenLabs API key to the fixed ElevenLabs service. No clear malware behavior, covert telemetry, credential harvesting beyond intended API-key use, persistence, or command injection is present. However, unvalidated effect names permit path traversal or absolute-path file operations, creating a meaningful security risk when the configuration is untrusted. FFmpeg executable paths are also environment-controlled. Validate effect names and enforce that resolved output paths remain within the output directory.

SUSPICIOUS. The skill’s media-editing, brand-scraping, transcription, audio generation, and QA capabilities mostly align with its stated purpose, and the ElevenLabs credential use matches that purpose. The main concerns are medium supply-chain risk from unpinned npx/transitive skill installs, broad .env discovery, and mandatory external delivery through another skill/review canvas whose exact endpoint and trust model are not fully shown. No confirmed malware or clear credential-harvesting behavior is present, but the external dependencies and outbound data flows are broader than a simple local editing skill.