paper-collage-ad

SUSPICIOUS. The core media-production workflow is largely coherent and proportionate, with mostly local rendering and explicit privacy rules. Risk comes mainly from the optional third-party IndexTTS-2 MLX installer and the handling of sensitive voice-cloning inputs without clear installer provenance in the skill text; this raises medium supply-chain and privacy concern, but there is no strong evidence of credential theft, hidden exfiltration, or confirmed malware.

No direct evidence of intentional malware/backdoors in this module. However, the code has significant security risks if an attacker can influence inputs or remote responses: (1) manifest-controlled scene.image can escape the intended directory via path traversal, enabling arbitrary local file disclosure that is then base64-uploaded to the remote API; and (2) the tool blindly fetches and writes the remote-provided video_url without validation, enabling SSRF/untrusted download and potentially unsafe processing of downloaded content by ffprobe. These issues are best treated as high-impact input-validation/trust-boundary gaps rather than proof of malicious payloads.

No direct malware behavior is evident in the script logic itself; it is a conventional bootstrapper that fetches and installs third-party runtime code and downloads model artifacts. The primary security concern is supply-chain integrity: it pulls/clones and installs remote code/dependencies without pinning or verifying versions/hashes/signatures, and it allows remote repository references to be redirected via environment variables. This warrants review/hardening (pin git commit/tags, pin uv dependency resolution with lock/constraints, verify model revisions/checksums, and optionally restrict/allowlist upstream sources).