web-video-presentation

SUSPICIOUS. The skill’s core capabilities fit its stated purpose as a web-video/TTS production workflow, and there is no clear malware or hidden exfiltration path. However, it expands trust to external TTS tooling, supports custom provider shell scripts, reads secrets from local environment/.env, and can redirect OpenAI credentials via OPENAI_BASE_URL. Those supply-chain and credential-forwarding behaviors are proportionate to the feature set but still medium risk.

This wrapper itself shows no standalone malicious payload, secrets, or direct exfiltration, but it creates a significant supply-chain and trust-boundary risk by dynamically `source`-loading a provider adapter chosen via user-controlled input. Additionally, chapter/step fields from audio-segments.json are used without sanitization to form filesystem paths under public/audio, which could enable path traversal or unintended writes if attacker-controlled inputs reach this runner. Overall safety depends on the integrity of the selected provider adapter and the trust level/sanitization of audio-segments.json contents.