← diffdock (scientific-agent-skills)
diffdock (scientific-agent-skills) — agentic threat model
This agent skill exposes a high-impact scientific model (DiffDock) that runs bundled Python/model code on the host, presenting significant local execution and code injection risks if integrated into an un-sandboxed agent framework.
OWASP AIVSS score rationale
| Autonomy of Action | 0.20 | |
| Goal-Driven Planning | 0.10 | |
| Self-Modification | 0.00 | |
| Dynamic Tool Use | 0.40 | |
| Persistent Memory | 0.00 | |
| Contextual Awareness | 0.30 | |
| Dynamic Identity | 0.00 | |
| Multi-Agent Interactions | 0.20 | |
| Non-Determinism | 0.50 | |
| Opacity & Reflexivity | 0.60 |
Scored with the canonical OWASP AIVSS formula (AIVSS calculator reference); agentic risk factors estimated from the agent’s described capabilities.
MAESTRO 7-layer threat model
Per-layer threats for this agent. Layers tagged “not certain from listing” are general, caveated commentary where the public description didn’t pin that layer.
Uses a specialized diffusion-based model (DiffDock) for molecular docking. Threats include adversarial input manipulation of protein/ligand structures to produce false binding predictions, or model evasion.
Not certain from the listing — relies on external structural biology files (PDB/SDF) as inputs. Gaps in data provenance or malicious structure files could lead to buffer overflows or parser exploits in underlying libraries.
The skill injects tool-specific guidance and runs Python/model code. Insecure tool integration or lack of input sanitization on the framework side could allow arbitrary code execution on the host.
Runs bundled Python/model code directly on the host. Without strict containerization or sandboxing, this poses a severe threat of host compromise and privilege escalation.
Not certain from the listing — no built-in logging, guardrails, or drift detection are mentioned for the execution of this specific molecular docking skill.
Not certain from the listing — being an open-source skill, it lacks explicit identity, authorization, or compliance controls, shifting all security responsibility to the deploying developer.
Designed as a single skill within a larger scientific agent-skills library. If integrated into multi-agent workflows, compromised upstream agents could feed malicious inputs to trigger execution vulnerabilities.
MAESTRO — the 7-layer agentic threat-modeling framework (Cloud Security Alliance / Ken Huang).