Back to skill

Security audit

Koyomi Behavior

Security checks for vulnerabilities and agentic risk

Overview

This is a behavior-guidance skill that promotes verification and cautious communication, with no evidence of code execution, exfiltration, persistence, or destructive actions.

Install this if you want the agent to be stricter about verifying work and stating evidence. Be aware that its trigger is broad and it may impose a Chinese/Kaguya-style persona or wording conventions in routine responses.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The activation condition is broad: it says to use this skill whenever the agent is asked to output a conclusion, complete a task, or make any judgment. That scope can cause the skill to trigger for ordinary interactions and override more specific task instructions, creating prompt-scope hijacking or unintended behavior changes across unrelated contexts. In this case the content is safety-oriented, which reduces severity, but the broad match remains a real control-plane weakness.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill imposes a fixed persona/style ('本小姐') and output convention without user choice or a clear operational necessity. While this is not directly a code-execution or data-exfiltration issue, it can interfere with user intent, locale expectations, or downstream automation that depends on neutral formatting, making the agent less predictable and harder to control. The surrounding skill is otherwise safety-focused, so the danger is limited but still real as a behavioral constraint vulnerability.

Content

No source excerpt is available for this finding.