ClawHub
Back to skill

Security audit

Discord 全能控制

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a disclosed Discord bot-control skill, but it mixes high-impact moderation with broad local exec/read/write claims and vague non-Discord automation scope.

Install only if you intend to let an agent operate a Discord bot with posting, deletion, role, and moderation powers. Keep moderation and role gates disabled unless needed, use a least-privilege bot token, require human confirmation for delete/kick/ban/timeout actions, and do not treat the vague command-whitelist language as an actual safety control.

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 (4)

Intent-Code Divergence

Medium
Confidence
92% confidence
Finding
The skill declares `exec` capability and later claims command execution is restricted to a whitelist, but the document never defines any actual command boundary, whitelist source, or enforcement mechanism. This mismatch can mislead operators into overtrusting the skill and approving workflows that may permit arbitrary or unintended command execution through the agent runtime.

Description-Behavior Mismatch

Medium
Confidence
88% confidence
Finding
The document expands from Discord bot control into generic file processing, command execution, and ops automation scenarios that are outside the stated scope. In a skill that already advertises `read`, `write`, and `exec`, this scope creep increases the chance that the agent or user will invoke broad local-system actions under the guise of a Discord skill.

Missing User Warnings

Medium
Confidence
84% confidence
Finding
The skill exposes destructive moderation actions such as timeout, kick, and ban without requiring an explicit confirmation or warning at the point of use. Because these actions directly affect users and community access, an agent could perform irreversible or high-impact moderation from ambiguous, mistaken, or prompt-injected instructions.

Missing User Warnings

Medium
Confidence
90% confidence
Finding
The example workflow deletes a message and times out a user as a normal recipe, but it does not include any caution about verification, human approval, or consequences to the affected user. Normalizing punitive actions in examples makes accidental or automated misuse more likely, especially in agent-driven environments where examples often become templates for execution.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.