ClawHub
Back to skill

Security audit

UniProt Query

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a legitimate UniProt lookup helper, with expected network queries and local report files but no evidence of hidden or destructive behavior.

Before running the examples, be comfortable with sending protein queries to UniProt and storing the returned metadata locally. Use a non-sensitive output directory, remove saved reports when no longer needed, and avoid placing sensitive research terms in queries if local retention or external API access is a concern.

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (2)

Description-Behavior Mismatch

Medium
Confidence
90% confidence
Finding
The example performs local file writes of query results and generated reports even though the skill is described as a query/retrieval capability. In agent environments, unexpected persistence expands the skill's privileges, can leak sensitive query contents or returned data to disk, and may violate least-privilege expectations for a read-only lookup tool.

Context-Inappropriate Capability

Medium
Confidence
93% confidence
Finding
A search-focused skill should not need filesystem write capability to fulfill its primary function, yet this code creates directories and writes JSON and text output locally. In a broader agent runtime, that unnecessary capability can be abused for unintended persistence, data staging, or storing sensitive biological queries/results where operators may not expect artifacts to remain.

Static analysis

No suspicious patterns detected.