fine-tuning-expert

Warn

Audited by Socket on Jul 15, 2026

1 alert found:

Anomaly
AnomalyLOW
references/deployment-optimization.md

No explicit malware or data-theft/backdoor behavior is evident in this fragment. However, it contains meaningful supply-chain code-execution surfaces: (1) trust_remote_code=True during model/vLLM loading can execute repository-provided custom code, and (2) export_to_gguf() runs external conversion/quantization scripts/binaries via subprocess.run() using paths influenced by LLAMA_CPP_PATH. If model repositories, adapters, and the llama.cpp toolchain directory are not tightly controlled and pinned, this module could be leveraged for arbitrary code execution during deployment/export. Hardening recommendations include disabling trust_remote_code when possible, pinning model/tool sources, and tightly controlling/sanitizing LLAMA_CPP_PATH and tool directories.

Confidence: 72%Severity: 50%
Audit Metadata
Analyzed At
Jul 15, 2026, 04:40 PM
Package URL
pkg:socket/skills-sh/neonetz%2Fopencode-skills%2Ffine-tuning-expert%2F@3889c3956e34641edc76eda31e66eef121211fa121f1ca0953d1de4702a2b757
Security Audit — socket — fine-tuning-expert