Ae1
High
- Category
- analysis-evasion
- Content
node scripts/run.mjs --sample # 先看内置干净样例
- Confidence
- 100% confidence
- Finding
- Referenced artifact was not completely inspected
Security audit
出口退税申报与单证核对（免费版）
Security checks for vulnerabilities and agentic risk
Overview
This is a local-only tax refund table checker with no evidence of hidden network access, persistence, or destructive behavior.
Use this only for local consistency checks on copied export tax refund tables, and do not treat its output as legal or tax authority guidance. Review the listed checks_not_run and have a qualified reviewer confirm tax-rate, classification, and filing decisions.
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
- 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 (7)
Ae1
High
- Category
- analysis-evasion
- Content
node scripts/run.mjs --sample # 先看内置干净样例
- Confidence
- 100% confidence
- Finding
- Referenced artifact was not completely inspected
Ae1
High
- Category
- analysis-evasion
- Content
node scripts/run.mjs --sample # 先看内置干净样例
- Confidence
- 100% confidence
- Finding
- Referenced artifact was not completely inspected
Description-Behavior Mismatch
Medium
- Confidence
- 98% confidence
- Finding
- The result metadata is internally inconsistent: `scope.checks` reports only the free checks, but `executed` is built from `CHECKS_GIVEN.concat(CHECKS_WITHHELD)`, which falsely indicates withheld checks were executed. In an audit/compliance skill, incorrect execution claims can mislead downstream users or orchestrators into trusting analysis that never ran, causing missed errors and unsafe business decisions.
Natural-Language Policy Violations
Medium
- Confidence
- 93% confidence
- Finding
- This code file contains user-facing help, usage text, errors, and status messages exclusively in Chinese. The file does not indicate that the skill is intentionally limited to Chinese-speaking users or provide any opt-in or alternative language, which can violate language/locale policy requirements.
Intent-Code Divergence
Low
- Confidence
- 84% confidence
- Finding
- The documentation says the tool does not determine which rebate-rate library/version applies and only checks internal consistency based on user-provided values. But `checkRateTier` emits a strong conclusion that the product code's applicable rebate rate differs and suggests expired library or misclassification, which goes beyond a purely neutral consistency statement. This is a mild contradiction between the stated non-judgment posture and the wording of actual findings.
Intent-Code Divergence
Low
- Confidence
- 95% confidence
- Finding
- The comment states that non-JSON input will be treated as plain-text materials such as '直接粘贴的合同全文', which contradicts the rest of the file's stated intent and usage that require an export tax refund declaration/document comparison table with headers. This is not just incomplete documentation; it describes a materially different input type than the skill claims to operate on.
Static analysis
No suspicious patterns detected.