ClawHub
Back to skill

Security audit

Saju Interpreter

Security checks for vulnerabilities and agentic risk

Overview

This skill is a small, disclosed Saju interpretation helper with an optional local calculator and no evidence of hidden data access, persistence, or destructive behavior.

Reasonable to install for structured Saju interpretation. If you use birth date/time conversion, verify the korean_lunar_calendar dependency and manually review 23:00-23:59 zi-hour cases or other school-specific calendar assumptions before relying on the result.

SkillSpector

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)

Description-Behavior Mismatch

Medium
Confidence
93% confidence
Finding
The script materially exceeds the skill's declared 'interpretation-only' scope by accepting calendar dates and deriving pillars from them. That mismatch is dangerous because downstream agents or users may rely on the manifest to decide whether the skill is appropriate, causing hidden behavior expansion and incorrect trust boundaries around date conversion logic.

Intent-Code Divergence

Medium
Confidence
90% confidence
Finding
The CLI exposes a `--zi-hour-mode` option and documents distinct handling, but the selected mode does not actually change the hour/day calculation logic; it only affects a note in the output. In a domain where 23:00-23:59 handling can alter the day pillar and therefore the full reading, this creates misleading behavior and silent correctness errors that users may not detect.

Static analysis

No suspicious patterns detected.