Security audit

Saju Interpreter

Security checks for vulnerabilities and agentic risk

Overview

This skill provides local Saju/Four Pillars interpretation guidance with a small local calculator, and I found no hidden, persistent, destructive, or data-exfiltrating behavior.

Before installing, treat the built-in pillar calculation as a convenience aid and verify important birth-chart derivations with a trusted 만세력 source, especially around lunar leap months and 자시 handling. The skill does not show signs of hidden access or persistence.

A.I.G

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

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 (6)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 99% confidence
Finding: The declared description says this skill should be used for interpretation of already-derived pillars and explicitly says to prefer it for interpretation, not calendar derivation, assuming the four pillars are already given unless calendar conversion is explicitly requested. The supplied code does the opposite primary task: it accepts calendar date/time inputs, converts solar or lunar dates into year/month/day pillars via KoreanLunarCalendar, computes the hour pillar, and returns structured pillar output. There is no code for interpreting 사주, 십성, 조후, 합충형해파, 신강약, 일간, 용신, or any reading logic from known pillars. This is a material description-behavior mismatch in primary purpose.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 91% confidence
Finding: The description says to use the skill when the user asks for a wide set of terms such as '사주', '명리', '팔자', or related discussion, but it does not provide exclusion examples or tighter context constraints. Some of these terms are broad enough to appear in general conversation, which could cause unintended activation.
Content: No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The manifest description limits the skill's primary role to interpreting known Four Pillars and says to assume the pillars are already given unless the user explicitly asks for calendar conversion. In contrast, the skill file's workflow explicitly tells the agent to use scripts/calculate-pillars.py when the user provides birth date/time, adding derivation behavior beyond the stated interpretation-first scope.
Content: No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The manifest describes this skill as an interpreter for already-derived Four Pillars and explicitly says to prefer it for interpretation, not calendar derivation unless the user explicitly asks for conversion. This script's primary behavior is to accept solar/lunar date and time, convert via KoreanLunarCalendar, and calculate/output pillars, which is a broader capability than the stated interpretation-focused scope.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The command-line interface presents '--zi-hour-mode' with choices 'civil' and 'split', implying the calculation behavior changes based on the selected mode. In practice, the hour pillar is always computed the same way, and 'split' only appends a note when hour == 23, so the documented/calimed behavior diverges from the actual computation.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Low

Category: Not specified by scanner
Confidence: 88% confidence
Finding: Earlier workflow guidance says that if the user gives birth date/time, the agent should use scripts/calculate-pillars.py for first-pass calculation. Later boundary guidance says that for full derivation from solar/lunar birth info, the user should be directed to a reliable 만세력 source first, which conflicts with the earlier prescribed behavior.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.