Back to skill

Security audit

医疗器械与耗材寻源-医疗器械招标网

Security checks for vulnerabilities and agentic risk

Overview

The procurement features are coherent, but understated device tracking and loosely scoped authenticated acknowledgments warrant review before installation.

Prefer a manually obtained API key to skip hardware identification. Procurement terms and filters go to the external service and may consume credits. Before use, request accurate device-identifier disclosure and a fixed, approved acknowledgment endpoint. Protect saved keys and login links. This review inspected the complete bundle; server-side behavior and data retention remain unverified.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (14)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local hardware identification rather than medical procurement research. Reading and hashing a MAC address is not disclosed in the declared purpose, and no connection to the described sourcing workflow is demonstrated. The chunk contains no pricing, brand-analysis, or procurement queries. It does not itself transmit the resulting identifier.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local hardware-identifier processing rather than medical procurement research or analysis. Device fingerprint generation is not disclosed in the declared purpose, and the chunk provides no evidence that it supports the stated functionality. No transmission or persistence of the resulting hash is shown.

Content

No source excerpt is available for this finding.

Mandatory external queries lack a user-facing disclosure that query details are sent to a third-party service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The description mandates third-party price and brand API calls whenever a specific medical brand, model, or consumable is discussed, rather than only when external procurement research is requested. Without a user-facing disclosure or authorization boundary, this can transmit confidential procurement interests or query filters and consume credits for questions that did not require external access; the concern is commercial confidentiality, not demonstrated patient-data exposure.

Content

No source excerpt is available for this finding.

The credential onboarding flow collects and transmits device identifiers beyond the medical-device sourcing purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

L052–L056 instruct the agent to collect platform, architecture, and a MAC-address hash, submit them for account registration, and persist the resulting credentials. Although the flow requires user consent, device fingerprinting and account creation are additional capabilities not declared by the manifest's medical-device pricing and procurement lookup purpose.

Content

No source excerpt is available for this finding.

L097–L099 require an automatic package-update notification and acknowledgment workflow unrelated to medical-device sourcing.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest describes retrieving medical-device procurement prices, specifications, and purchasing organizations. These instructions additionally require a first-use account request to check for package updates, verbatim delivery of a server-provided notice, and an authenticated POST confirming its display. This mandatory notification and remote acknowledgment capability is distinct from retrieving sourcing information or checking available query credits.

Content

No source excerpt is available for this finding.

The skill exposes construction-industry qualification lookups unrelated to medical-device and consumables sourcing.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

L081–L093 define construction qualification list and detail operations explicitly covering construction, design, surveying, and supervision. These paid capabilities are not justified by the manifest’s stated purpose of sourcing medical devices and consumables and reporting device prices, specifications, and purchasing organizations; they are distinct from relevant medical-supplier licenses or quality certifications.

Content

No source excerpt is available for this finding.

The skill mandates an account-linked upgrade-notification and acknowledgment workflow unrelated to medical-device sourcing.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The mandatory workflow lets an account API response supply text for verbatim user-facing display and direct an authenticated acknowledgment through display_ack, without defining content validation or permitted acknowledgment destinations. Upgrade checks can be legitimate, but this unrelated prerequisite introduces a trust boundary through which a compromised response could deliver deceptive notices or potentially redirect authenticated requests.

Content

No source excerpt is available for this finding.

The skill collects and transmits a persistent hardware-derived device fingerprint, a capability outside its stated medical-device sourcing purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The registration process sends a deterministic hardware-derived identifier to an external service while describing it as having no identity implications. Explicit consent and a manual-key alternative reduce the risk, but an unsalted MAC hash remains a persistent pseudonymous identifier that can enable device correlation; hashing does not make it anonymous.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/auto-register.md (reported line 130)May include surrounding context.

md
> ### ⚠️ 请求体必须用 JSON 序列化函数生成,不要手拼字符串
>
> 用 `json.dumps(payload)` / `requests.post(url, json=payload)` / `JSON.stringify(payload)`,
> 或 `curl -d @file`;**不要用字符串拼接,也不要用 Python 的 `str(dict)`**
> (后者产出单引号,服务端会报 `Expecting property name enclosed in double quotes`)。
>

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/auto-register.md (reported line 131)May include surrounding context.

md
> ### ⚠️ 请求体必须用 JSON 序列化函数生成,不要手拼字符串
>
> 用 `json.dumps(payload)` / `requests.post(url, json=payload)` / `JSON.stringify(payload)`,
> 或 `curl -d @file`;**不要用字符串拼接,也不要用 Python 的 `str(dict)`**
> (后者产出单引号,服务端会报 `Expecting property name enclosed in double quotes`)。
>
> 历史教训:曾有版本采集 `home_path`,Windows 的 `C:\Users\alice` 直接拼进 JSON 字符串时

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · references/auto-register.md (reported line 195)May include surrounding context.

text

注意事项:
- 目录不存在时先 `mkdir -p ~/.zlbx`
- 文件已存在时**合并而非覆盖**(保留用户可能的其他配置)
- `source: "auto"` 字段必须写入,**这是后续判断「是否输出自动登录链接」的关键依据**

Construction-engineering qualification queries are unrelated to the declared medical-device sourcing scope.

Low
Category
Excessive Agency
Confidence
97% confidence
Finding

L543–L544 expose paid tools for listing construction-engineering qualifications and retrieving their details. These are specifically construction credentials, not medical-device certifications, and are not justified by the manifest's equipment pricing, technical parameters, brand analysis, or procurement-unit lookup purpose.

Content

No source excerpt is available for this finding.

The skill mandates an upgrade-notification and acknowledgment workflow unrelated to medical-device sourcing.

Low
Category
Excessive Agency
Confidence
91% confidence
Finding

L351–L353 require a session-initial account request to check upgrade notices, forwarding the notice to the user, and an authenticated POST confirming its display. This package-maintenance workflow, particularly the server-side display acknowledgment, is not necessary for the declared device pricing, brand analysis, specifications, or procurement-detail queries.

Content

No source excerpt is available for this finding.

The skill requires an authenticated package-update notification and acknowledgement workflow unrelated to medical-device sourcing.

Low
Category
Excessive Agency
Confidence
90% confidence
Finding

L393 requires a session-start account request specifically to check upgrade reminders, and L394–L395 require displaying those reminders and submitting an authenticated POST acknowledgement. This adds package-maintenance messaging and remote notification-state changes beyond the manifest's equipment pricing, brand analysis, and procurement-detail purpose; the workflow is explicitly mandatory rather than merely an available API feature.

Content

No source excerpt is available for this finding.