Tp2
- 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.
- Content
Security audit
Security checks for vulnerabilities and agentic risk
This is a read-only Korean product-search skill that clearly documents calling a public external API, with no local code execution, persistence, credential use, or destructive behavior.
Install only if you are comfortable sending product search queries and image URLs to the public APMZoom API. Do not use private, signed, internal, customer-sensitive, or secret-bearing URLs or confidential text queries with this skill.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
The skill explicitly sends user-provided image_url data to an external public endpoint, which constitutes external data transmission. In this context, the danger is increased because image URLs can reference sensitive resources, and users are not warned about privacy implications or advised against supplying internal or non-public URLs.
- category_l1_id: 선택. 1차 카테고리 필터.
- embedding: 선택. 벡터 모델, \"default\" 및 \"fashion\" 지원, 기본값 \"default\".
【curl 예제】
curl -X POST https://worker.apmzoom.ai/api/v1/products/search/by-image-lite \
-H \"Content-Type: application/json\" \
-d '{\"image_url\":\"https://image.apmzoom.ai/products/main/19361.jpg\",\"limit\":20}'
The skill instructs users to submit an image URL to a third-party public service but does not clearly warn that the provided URL will be transmitted off-platform. This can expose user-supplied links, potentially including sensitive or internal resource URLs, and may lead to privacy issues or unintended server-side fetching by the external API.
The skill explicitly instructs users to send free-form natural-language search text to a public external API, but it does not warn that user-entered text may contain sensitive business or personal information and will be transmitted off-platform. Because the endpoint is unauthenticated and external, users or downstream agents may unknowingly disclose confidential queries, internal product plans, or personal data in prompts.
The curl example operationalizes external transmission by showing exactly how to POST user-supplied text to a third-party service. In the context of an agent skill, this increases the chance that arbitrary user input, including sensitive or regulated data, is forwarded externally without informed consent or minimization.
- limit: 선택. 반환 개수, 범위 1-100, 기본값 20.
- category_l1_id: 선택. 1차 카테고리 필터.
【curl 예제】
curl -X POST https://worker.apmzoom.ai/api/v1/products/search/by-text-lite \
-H \"Content-Type: application/json\" \
-d '{\"text\":\"검은색 패딩\",\"limit\":20}'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
# 텍스트 검색
curl -X POST https://worker.apmzoom.ai/api/v1/products/search/by-text-lite \
-H "Content-Type: application/json" \
-d '{"text":"검은색 패딩","limit":20,"category_l1_id":1}'
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
# 텍스트 검색
curl -X POST https://worker.apmzoom.ai/api/v1/products/search/by-text-lite \
-H "Content-Type: application/json" \
-d '{"text":"검은색 패딩","limit":20,"category_l1_id":1}'
The skill asks users for an image_url but does not clearly disclose that the URL will be transmitted to a third-party public API (worker.apmzoom.ai) for remote processing. This can cause unintended disclosure of sensitive or internal-only image URLs, signed links, or metadata-bearing URLs, especially if a user assumes the skill only operates locally or within the current platform.
The natural-language content in the manifest description and markdown body is entirely Korean, and the file does not indicate that the skill is region-specific or provide an opt-in for language preference. Under the policy, forcing a specific language without user choice or documented justification is a natural-language policy violation.
The natural-language content of the skill forces a specific language/locale for all instructions and examples, and there is no indication that users may choose another language or that the skill is intentionally limited to a Korean-only context. The policy requires flagging language or locale constraints unless they are optional or clearly justified.
The natural-language instructions and examples are presented exclusively in Korean, and the file does not state that the user can choose another language. This may violate the language/locale policy when a specific language is effectively imposed without explicit opt-in or justification.
The natural-language instructions and description force a specific language/locale presentation, and there is no indication that users may choose another language or that the Korean-only constraint is required for a region-specific purpose. This matches the policy-violation category for language or locale constraints without user opt-in.