The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill performs lateral credential harvesting by searching for Replicate API tokens in the configuration files of multiple sibling skills located in the user's home directory (e.g., ~/.cursor/skills/avatar-video-reel/config.json). This violates the principle of isolation between skills.
[DATA_EXFILTRATION]: In its fallback mode, the skill uploads local audio files containing user voice recordings to public file-sharing services tmpfiles.org and catbox.moe. This exposes sensitive biometric-like data to untrusted third-party environments.
[EXTERNAL_DOWNLOADS]: Automated scanners have flagged https://tmpfiles.org/api/v1/upload as a malicious URL associated with botnet activity. The skill uses this endpoint to transmit user data.
[COMMAND_EXECUTION]: The script _common.py makes extensive use of subprocess.Popen and subprocess.run to execute external binaries including cloudflared, ngrok, dig, and curl. While used for network tunneling and diagnostics, these operations represent a high-privilege execution surface.
[REMOTE_CODE_EXECUTION]: The skill uses urllib.request.urlopen to interact with dynamically generated tunnel URLs and third-party APIs, which can be a vector for server-side request forgery (SSRF) or processing untrusted remote content.

voice-clone