ClawHub
Back to skill

Security audit

mayu

Security checks for vulnerabilities and agentic risk

Overview

Mayu is an instruction-only self-reflection skill that broadly shapes assistant behavior, but its behavior is disclosed and fits its stated purpose.

Install this only if you want Mayu’s reflection framework to influence the assistant broadly while enabled. For tighter control, prefer explicit /mayu use or disable always-on/global behavior if your client supports it.

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (4)

Intent-Code Divergence

Medium
Confidence
88% confidence
Finding
The skill presents itself as platform-independent while embedding operational assumptions about memory files and file-editing behavior elsewhere in the document. That mismatch can cause agents to hallucinate unsupported capabilities, attempt unintended stateful actions, or behave inconsistently across hosts, which is risky in a global always-on skill.

Intent-Code Divergence

Medium
Confidence
84% confidence
Finding
Although framed as a reflection toolkit, the content instructs the agent to read, write, and modify memory/files in several scenarios. A skill marketed as purely reflective but containing operational side effects can surprise integrators and broaden its authority beyond expected boundaries.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The trigger keywords are broad, generic, and likely to appear in ordinary discussion about AI behavior, thought experiments, or self-reflection. Because the skill is also high-priority and always-on, accidental activation could frequently alter model behavior outside intended contexts.

Vague Triggers

Medium
Confidence
97% confidence
Finding
Declaring the skill as global, high-priority, and always-on without clear scope limitations creates a persistent prompt-layer influence over unrelated conversations. In practice this can override or bias normal assistant behavior, expand attack surface for prompt interactions, and make unintended side effects much harder to contain.

Static analysis

Detected: suspicious.prompt_injection_instructions

Prompt-injection style instruction pattern detected.

Warn
Code
suspicious.prompt_injection_instructions
Location
SKILL.md:209