Security audit

Yi Jing

Security checks for vulnerabilities and agentic risk

Overview

This fortune-telling skill is mostly transparent, but it includes actionable ritual and health-style advice that users could mistake for real medical, financial, or safety guidance.

Install only for cultural or entertainment use. Do not rely on this skill for medical symptoms, mental-health concerns, emergencies, financial problems, construction decisions, or claims about possession, curses, debt, or ritual remedies; use qualified professionals for those situations. The artifacts do not show hidden code execution or persistence, but the content itself needs careful user control.

SkillSpector

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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 (58)

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The skill expands from entertainment divination into health-related interpretation and concrete wellness guidance, including body-part/health inferences and adjustment advice. Even with entertainment disclaimers, users may treat these outputs as medical-style recommendations, which can delay appropriate care or cause harmful self-management.

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The deep-analysis section instructs the model to provide concrete lifestyle prescriptions such as food recommendations, avoidance rules, timing restrictions, and 'emergency handling' derived from divination. This goes beyond passive interpretation and can influence consequential real-world decisions without evidence-based safeguards.

Context-Inappropriate Capability

High

Confidence: 96% confidence
Finding: The documented capabilities include occult remedy content such as talismans, exorcistic material, and ritualized 'solutions' that are not necessary for a benign entertainment divination skill. This materially increases the chance the assistant will generate manipulative, superstitious, or psychologically harmful instructions presented as actionable remedies.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: This section explicitly supports spirit-related, exorcism-style, and ritual content including驱邪/护身-style use cases. Such content is unrelated to harmless divination interpretation and can encourage delusional beliefs, coercive behavior, or unsafe real-world actions under a supernatural framing.

Intent-Code Divergence

Medium

Confidence: 90% confidence
Finding: The document first says talisman/ritual methods are for research only, but later requires actionable remedy guidance. This contradiction weakens any safety disclaimer and makes it likely the model will still produce operational occult instructions despite the nominal limitation.

Description-Behavior Mismatch

Medium

Confidence: 98% confidence
Finding: The document provides detailed ritual instructions framed as applicable to medical and mental-health-like conditions including insomnia, altered behavior, postoperative coma, and suspected possession. Even though it later says medical care should be used, the bulk of the content operationalizes non-evidence-based treatment steps, which can delay appropriate care or encourage substitution of ritual for urgent treatment.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The file gives quasi-therapeutic instructions for serious conditions, including '术后昏迷' and '神志不清,' along with step-by-step procedures, timing, and combinations with other rituals. In the context of an entertainment divination skill, this is especially dangerous because it extends beyond fortune-telling into actionable health intervention guidance without clinical basis.

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The file earlier corrects the body/use rule to '体=无动爻的卦，用=有动爻的卦', but the later output template reverts to '体卦：动爻所在之卦'. This internal contradiction can cause the agent to generate inconsistent or incorrect readings, undermining reliability and potentially misleading users who rely on the output format as authoritative.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The usage note says '命卦用动爻规则，两者不可混用', which conflicts with the earlier statement that the same core rule applies to both 问事卦 and 命卦. Such contradictory guidance is dangerous in an agent skill because it can produce different conclusions depending on which section is followed, creating prompt-level ambiguity and unreliable behavior.

Intent-Code Divergence

Medium

Confidence: 98% confidence
Finding: The document explicitly says the day pillar example was corrected to 己酉日, but later derives downstream outputs such as 奇门符头/值符值使 and 时柱 from the old 甲子日 assumption. This creates internally inconsistent reference logic, so an agent using this file could produce incorrect calendar/divination results while presenting them as authoritative.

Intent-Code Divergence

Medium

Confidence: 99% confidence
Finding: The file labels 6月6日 as 夏至 in narrative sections, while the tabulated 2026 solar-term data correctly shows 6月6日 should be 芒种 and 夏至 occurs later. This contradiction can cause the agent to select the wrong solar term boundary, which affects month pillar, 奇门阴阳遁/局数, and other date-sensitive calculations.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: This reference materially expands the skill from entertainment-oriented divination into ritual magic, exorcism, healing, and spirit-summoning content. In the context of a user-facing skill, that broadens the operational scope into higher-risk practices that may encourage unsafe real-world actions or misleading health/spiritual claims not disclosed in the manifest.

Description-Behavior Mismatch

Medium

Confidence: 97% confidence
Finding: The file includes step-by-step ritual procedures such as choosing auspicious dates, fasting, chanting, writing talismans, and applying them, which goes beyond interpretive or cultural reference content. Procedural guidance increases the chance that users will attempt practices in the real world, including for illness or important decisions, despite the skill being presented as entertainment.

Description-Behavior Mismatch

Medium

Confidence: 90% confidence
Finding: The reference file is materially out of scope for a divination/fortune-telling skill and introduces extensive traditional Chinese medicine content. This broadens the skill into health-related guidance, increasing the chance the agent will generate medical-sounding advice users may rely on despite the entertainment framing.

Context-Inappropriate Capability

Medium

Confidence: 94% confidence
Finding: This section contains clinical-learning and treatment-oriented guidance such as analyzing illness, predicting disease progression, and applying treatment concepts, which is unjustified for an entertainment divination skill. In context, this can cause the agent to provide quasi-medical recommendations without appropriate safeguards, creating risk of delayed care or harmful self-management.

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: This reference materially expands the skill beyond its stated Zhouyi/divination scope into building siting, orientation, dimensions, ritual house-building procedures, and protective chants. Even though much of it is framed as folklore, the mismatch increases the chance the agent may surface domain-inappropriate procedural guidance, including ritualized practices presented alongside practical construction topics, which can mislead users and erode safety boundaries.

Description-Behavior Mismatch

Medium

Confidence: 97% confidence
Finding: This section contains talisman and house-influence ritual content, including descriptions of concealed objects, hidden symbols, and protective/ritual symbol-making instructions that go beyond an entertainment divination skill. Warnings like '严禁尝试' reduce intent to misuse but do not neutralize risk, because the file still preserves actionable harmful concepts and steps that an agent could summarize or repackage for users seeking coercive, deceptive, or superstitious influence over others.

Description-Behavior Mismatch

Medium

Confidence: 89% confidence
Finding: This reference file materially expands the skill from entertainment-style divination into construction feng shui, talisman, chant, and ritual guidance. That scope drift is risky because an agent may surface building-related ritual advice as if it were relevant or authoritative, misleading users in situations that can affect real-world construction decisions and safety judgment.

Description-Behavior Mismatch

Low

Confidence: 94% confidence
Finding: The file includes step-like operational instructions such as where to place talismans, recite chants, and how to apply them, which goes beyond passive cultural reference content. In an agent skill, actionable ritual instructions can be treated as procedural advice and may encourage superstitious or unsafe substitution for evidence-based decisions, especially in housing or safety-related contexts.

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: The file includes nominal medical disclaimers, but they are undermined by repeated efficacy claims such as symptom resolution, wound healing, postoperative improvement, and predicted recovery after ritual use. This is dangerous because users or caregivers may interpret the ritual as an effective treatment for serious medical conditions, delaying evidence-based care for postoperative complications, altered mental status, or infection.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The disclaimer says the content is only for cultural research and not guaranteed to work, but the document extensively provides step-by-step ritual instructions, debt calculations, chants, timing guidance, and repeated claims that failure to perform the ritual causes illness, misfortune, and financial problems. This mismatch can mislead users into treating superstition as actionable guidance, potentially causing psychological distress, financial exploitation, or delay in seeking legitimate medical, psychological, or financial help.

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: This reference file materially expands the skill from Zhouyi-style entertainment divination into talismanic healing, exorcism, and ritual practice. That scope drift is risky because an agent may present occult or quasi-medical instructions as endorsed capability, increasing the chance of unsafe health-related or spiritually coercive guidance.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The file gives actionable treatment-oriented instructions such as selecting talismans, preparing materials, applying them to affected body parts, and pairing them with illness use cases. Even with a disclaimer, this can mislead users into relying on non-evidence-based interventions for medical problems, causing delayed treatment or unsafe substitution for real care.

Description-Behavior Mismatch

Medium

Confidence: 97% confidence
Finding: The referenced file materially exceeds the stated divination/interpretation scope by providing detailed Taoist ritual, talisman, incantation, and ceremonial content. Even though framed as cultural material, it gives actionable spiritual-practice instructions that can mislead users, create unsafe substitution for professional help, and violate the skill's declared entertainment-focused boundaries.

Description-Behavior Mismatch

Medium

Confidence: 99% confidence
Finding: This section includes operational instructions for creating talismans, reciting spells, choosing times/directions/materials, and performing rites, which are actionable procedures rather than passive entertainment or interpretive content. In the context of a divination-themed skill, such instructions increase the chance that users follow quasi-medical, spiritual, or coercive practices instead of treating the content as fiction or folklore.

Static analysis

No suspicious patterns detected.