Security audit

大藏经CBETA2026

Security checks for vulnerabilities and agentic risk

Overview

This is a documentation-only CBETA scripture lookup skill with disclosed MCP knowledge-base use and no hidden code, persistence, or destructive behavior.

Installers should understand that this skill is meant for CBETA-based Buddhist text lookup and citation checking. Confirm before allowing it to edit documents through Tencent Docs or local docx tooling, and treat its Simplified Chinese corpus as a disclosed coverage limitation.

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 (5)

Tp2

High

Category: MCP Tool Poisoning
Confidence: 85% confidence
Finding: Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The trigger list includes many broad, common Buddhist-domain terms, which can cause the skill to activate in conversations that only tangentially mention Buddhism. Over-broad invocation increases the chance of context hijacking, unnecessary tool use, and the model being steered into this skill when a narrower or safer skill would be more appropriate.

Natural-Language Policy Violations

Medium

Confidence: 87% confidence
Finding: The file repeatedly describes the skill as a “简体版” resource and presents that locale choice as mandatory behavior, but does not offer the user a language/script choice or explain a justified locale restriction. This can violate language/locale policy when users may prefer or require Traditional Chinese for Buddhist source work.

Description-Behavior Mismatch

Medium

Confidence: 84% confidence
Finding: The manifest and overview frame the skill as a precise CBETA scripture retrieval and quotation-verification tool using a single subscribed knowledge base. However, the deep-research section says to extract '古代注释 + 近现代学者观点', which goes beyond scripture/CBETA retrieval and citation checking into generalized research synthesis not justified by the stated single-source purpose.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The document states the skill uses the subscribed CBETA knowledge base as its only data source and does not automatically search the web. Elsewhere it instructs use of '近现代学者观点', CBETA online link generation, and even recommends other tools such as python-docx and Tencent Docs, which contradict the strict 'only data source' framing rather than merely omitting details.

Static analysis

No suspicious patterns detected.