paper-collage-ad

This is primarily an API client for uploading locally referenced images and prompts to a remote generation service, polling for completion, and downloading the resulting video. It does not show classic malware behavior in the Node.js logic, but it has significant security risks if either the manifest or remote responses are attacker-influenced: (1) manifest-driven arbitrary local file read due to lack of path confinement for scene.image, followed by exfiltration of those bytes to the configured ARK_BASE; and (2) downloading a remote-provided video_url without validating destination or content constraints. These issues make the code potentially dangerous in untrusted supply-chain scenarios, even though it is not overtly malicious by itself.

No strong indicators of intentional malware within this module (no obfuscation, no credential harvesting beyond using the provided API key for its intended TTS function, no persistence or reverse-shell behaviors). However, the module has a concrete security weakness: `manifest.lines[].id` is used directly to form output paths and cleanup paths (MP3/WAV filenames) without sanitization, enabling path traversal/overwrite/deletion of files outside the intended --outdir in environments where such traversal is possible. Additionally, the script processes remote-generated media with ffmpeg/ffprobe, which increases exposure to media parsing risks. Treat the manifest as untrusted input and validate/sanitize line.id and response data to reduce risk.

This is a setup/installer script that performs substantial supply-chain operations: it fetches a third-party runtime repo (clone/pull without commit pinning or signature/hash verification) and then runs uv sync/install based on that fetched code, plus it downloads a model from Hugging Face without explicit integrity verification. No direct malware behavior is visible in the script itself, but the overall security posture is meaningfully exposed to upstream/registry compromise and environment manipulation. Recommend pinning runtime/model revisions and adding checksum/signature verification for downloaded artifacts before execution/installation.