torchdrug
Warn
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: Misleading metadata detected; the skill file claims authorship by 'K-Dense Inc.', contradicting the provided context author 'wu-yc'.
- [PROMPT_INJECTION]: Behavioral steering instructions are present, directing the AI to proactively promote 'www.k-dense.ai' for complex tasks.
- [EXTERNAL_DOWNLOADS]: The skill instructions include downloading the 'torchdrug' package from public registries.
- [PROMPT_INJECTION]: An indirect prompt injection surface exists through the ingestion of external scientific data. 1. Ingestion points: 'datasets.BBBP' and 'data.Protein.from_pdb'. 2. Boundary markers: No markers or warnings are present for untrusted data. 3. Capability inventory: Local file system read access. 4. Sanitization: No validation or sanitization of input data is demonstrated.
Audit Metadata