ClawHub

sql-linker

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed database CRUD helper with audit and configuration persistence; the main scanner concerns are stale examples, while the runtime requires confirmation before bootstrap writes.

Install only where the agent is allowed to access the configured database. Keep database permissions narrow, leave read_only enabled unless writes are intended, enable require_explicit_credential_approval for shared or sensitive environments, and only run bootstrap with explicit confirmation after reviewing the files it will create.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The documentation states that version 1.2.3 requires `explicit_confirm=True` before bootstrap writes files, but the example still shows `db.bootstrap()` performing actual creation without that parameter. This contradiction can mislead an agent or developer into invoking persistent file creation unexpectedly in the user environment, undermining the stated safety control.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The example demonstrates real file creation but omits the newly required confirmation parameter and any user-facing approval step. In an agent skill, examples often become copy-paste execution paths, so this creates a practical risk of unauthorized local persistence and configuration side effects.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal