Back to skill
Skillv1.2.0
VirusTotal security
Doc-to-LoRA · External malware reputation and Code Insight signals for this exact artifact hash.
Scanner verdict
SuspiciousApr 30, 2026, 5:37 AM
- Hash
- c6d2b1d41eccf8b3b3cd53bacb56cbea5a27be628b7eafb4bf9b689636a0bcc7
- Source
- palm
- Verdict
- suspicious
- Code Insight
- Type: OpenClaw Skill Name: doc-to-lora-hyper Version: 1.2.0 The skill implements a legitimate document internalization technique but contains a high-risk security vulnerability. Specifically, `scripts/internalize.py` uses `torch.load(weights_only=False)`, which allows for arbitrary code execution (RCE) via Python's pickle module if a malicious checkpoint is provided. While the documentation in `SKILL.md` and the script itself explicitly warn users to only load trusted checkpoints from the official SakanaAI HuggingFace repository, the inherent risk remains. Additionally, `scripts/setup.sh` executes an external script (`install_mac.sh`) not included in the bundle and requires a sensitive `HF_TOKEN` for operation.
- External report
- View on VirusTotal