ClawHub
Back to skill

Security audit

ASI-Proxy Phase Skill

Security checks across malware telemetry and agentic risk

Overview

The skill’s actions are mostly disclosed and purpose-aligned, but the inspected package appears to be missing a bundled paper index needed for normal operation.

Install only if you want a research/intervention workflow that can inspect and modify a declared workspace when you ask it to. Run network maintenance checks only after approving public GitHub/Hugging Face access, and be aware that GitHub tokens in the environment may be used for rate limits. Also verify the package contents because this inspected copy fails its own doctor check due to a missing references/papers.jsonl file.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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
Findings (3)

Lp3

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding
The skill instructs the agent to run scripts that read and write files, inspect environment state, and perform source-refresh checks, yet no permissions are explicitly declared. That mismatch weakens policy enforcement because operators may trust the manifest while the workflow still drives privileged behavior such as workspace mutation and potential network use.

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding
The skill is presented as a diagnosis/improvement framework, but its documented behavior includes operational tooling such as installation checks, index rebuilding, drift checks against public services, packet generation, and workspace mutation. This description-behavior gap can mislead users and automated policy systems into invoking a skill with broader operational effects than expected, increasing the risk of unauthorized network access or file changes.

Vague Triggers

Low
Confidence
88% confidence
Finding
The approval-required action list includes the term "external communication" without defining what counts as external, which creates policy ambiguity at an authority boundary. In an agent skill, unclear approval triggers can be exploited or accidentally interpreted narrowly, allowing outbound messaging, API calls, issue creation, or other off-workspace disclosures to occur without escalation.

VirusTotal

52/52 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.