Pharmaclaw Alphafold Agent

• Outcome: Review
• Latest audit: May 29, 2026, 9:24 PM
• Version:
  1.0.0

Overview

The skill does not show theft or persistence, but it presents placeholder scientific calculations as usable AlphaFold, pocket detection, and docking results.

Install only for experimentation, demos, or code review. Do not rely on its structures, pockets, docking scores, or visualizations for research, medicinal chemistry, IP, or safety decisions unless the placeholder methods are replaced with validated tools and outputs are clearly labeled with method provenance.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Intent-Code Divergence

High

• Confidence: 97% confidence
• Finding: The skill claims to perform ESMFold prediction but actually writes a hardcoded mock PDB file. In a scientific or drug-discovery workflow, this can mislead downstream agents or users into making decisions based on fabricated structural data, creating integrity and safety risks even without code execution.

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The binding-site routine ignores the parsed structure and returns a fixed pocket every time, despite being presented as real pocket detection. This is dangerous because downstream medicinal chemistry, docking, or prioritization steps may trust false active-site information and produce unsafe or invalid conclusions.

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The docking function is represented as ligand docking, but it only embeds a ligand and computes a pseudo-score from molecular weight. In a pharmacology context, presenting this as docking can materially deceive downstream systems or users about binding affinity and lead to unsafe scientific decisions.

Vague Triggers

Medium

• Confidence: 87% confidence
• Finding: The trigger list includes very generic terms such as 'fold', 'structure', 'target', and 'binding site', which increases the chance of accidental invocation in unrelated conversations. Because this skill can make network requests and write local artifacts, unintended activation broadens exposure and can trigger unnecessary external calls or file creation without clear user intent.

VirusTotal

63/63 vendors flagged this skill as clean.