Voice.ai: Creator Voiceover Forge

Turn scripts into publishable voiceovers with Voice.ai TTS, including segments, chapters, captions, and video muxing.

What to check before installing or running: - Confirm the API key requirement: SKILL.md and the code require VOICE_AI_API_KEY (or VOICEAI_API_KEY). The registry metadata incorrectly lists no required env vars — treat the SKILL.md/code as authoritative. - Do not paste a production VOICE_AI_API_KEY until you verify the service endpoint and publisher. The code points to https://dev.voice.ai and TROUBLESHOOTING warns that production endpoints may be placeholder; use --mock to test locally without sending data. - Verify the publisher/source: homepage is missing and owner id is an opaque string. README references a GitHub repo (gizmoGremlin) — inspect that upstream repo or contact the author to confirm authenticity before trusting real credentials. - Review bundled binary (voiceai-vo.cjs) or run in an isolated environment/container. Running with --mock first lets you exercise the pipeline without network calls. - Legal/privacy note: the voice catalog includes names that imply celebrity/character voices; ensure you’re comfortable with any potential voice-mimicry/licensing issues for your use case. If you want higher confidence: ask the publisher for a canonical homepage or GitHub link, verify the API base (production vs dev), and run the bundled CLI in --mock mode to validate local behavior before providing secrets.

Type: OpenClaw Skill Name: voiceai-voiceover-creator Version: 0.1.3 The skill is classified as suspicious due to multiple shell injection vulnerabilities in `src/ffmpeg.ts`. User-controlled input paths (e.g., `--video`, `--audio`, `--out`) are directly interpolated into commands executed via `child_process.execFile` and, more critically, embedded into generated shell scripts (`replace-audio.sh`, `replace-audio.ps1`). An attacker could craft malicious input (e.g., `--video 'foo.mp4; rm -rf /'`) which would be executed if the user runs the generated scripts, or potentially via direct `ffmpeg` execution depending on argument parsing. There is no evidence of intentional malice, but the lack of input sanitization creates a severe remote code execution risk.