T09 · Insecure Skill Coding Practices
- Location
SKILL.md:278- Finding
Misleading Local-Only Processing Claim Conflicts with Third-Party API Disclosure
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, lines 141-143 and 278
Vulnerability Type: Inaccurate privacy boundary and external data-processing disclosure
Risk Level: MediumVulnerable Snippets
SKILL.md:141-143:markdown 1. **数据查询**：通过云旅AI MatchGPT API获取广交会展商数据 2. **信息补全**：自动补全联系方式、社交媒体、决策人信息 3. **匹配评分**：MatchGPT从产品匹配度、采购规模、地理分布、合作潜力4个维度评分SKILL.md:278:markdown - **本地处理**：查询条件和中转数据仅在本地处理The first excerpt instructs the Skill to obtain exhibitor data and perform contact enrichment through the Yunlv AI MatchGPT service. The second excerpt claims that query conditions and intermediate data are processed only locally.
Technical Analysis
The documented workflow depends on the external endpoint
https://api.yunlvai.comand uses query criteria to retrieve and enrich exhibitor information. At least some request data must therefore cross the local trust boundary. Depending on the API implementation, transmitted information may include product criteria, target markets, company names, booth identifiers, lead records, or contact-enrichment inputs.This conflicts with the unconditional claim that query conditions and intermediate data are processed only locally. The package does not document the exact request schema, data minimization controls, provider retention period, subprocessors, or whether user consent is obtained before each transmission. No executable implementation is included, so encryption behavior, deletion guarantees, and the actual fields sent cannot be independently verified.
The issue is an insecure Skill configuration and documentation practice because users may make disclosure decisions based on an inaccurate description of the data boundary. It is not evidence that the declared API provider is malicious.
Attack Path
- A user invokes the Skill and supplies business-sensitive search criteria, a company name, booth information, or contact-related context ...[truncated 1467 chars]
- Remediation
View remediation
Remediation Suggestions
- Replace the local-only statement with an accurate data-flow disclosure specifying that selected query information is transmitted to
https://api.yunlvai.com. - Document the exact fields sent to the service, including whether company names, contact details, free-form prompts, or prior lead records are included.
- Require explicit user confirmation before transmitting business-sensitive or personal data to the external API.
- Apply data minimization by sending only fields strictly required for each query and redacting unrelated user context.
- Document transport security, provider retention periods, deletion procedures, processing regions, subprocessors, and applicable privacy terms.
- Avoid sending existing contact records for enrichment unless the user has authorized that specific disclosure and has a lawful basis for processing.
- Add implementation-level controls that prevent API keys, request bodies, contact data, and API responses from being written to logs.
- Provide a local-only mode that disables external enrichment when the user does not consent to third-party processing.
- Add tests or auditable request logging with sensitive fields redacted to verify that implementation behavior matches the published privacy statement.
- Replace the local-only statement with an accurate data-flow disclosure specifying that selected query information is transmitted to