smolvlm

[Skill Scanner] Installation of third-party script detected The provided file is a benign-sounding README for a local vision-language skill that runs a SmolVLM model. I found no explicit malicious code or data-exfiltration instructions in the text itself. The primary security concern is supply-chain and operational: an automatic large (~4GB) model download from an unspecified/unverified source and an incorrect/misleading install command increase risk. Recommendations: provide authoritative model download URLs, cryptographic checksums or signatures, a corrected and minimal install command, documentation of where model files are stored, affirm whether downloads use TLS and whether the package verifies signatures, and warn users that images may contain sensitive data which will be output to logs/console. LLM verification: No explicit malicious code is visible in the provided README fragment. The primary security concern is supply-chain risk: an unpinned pip install and an unspecified first-run model download (~4GB) without checksums or documented endpoints. These gaps enable an attacker who can compromise the package index or model hosting to deliver malicious code or model weights that run locally. Mitigations: pin package versions and hashes, publish and verify model checksums/signatures, document network endpo