Back to skill

Security audit

全国采购与招标信息总库-中国采购与招标网

Security checks for vulnerabilities and agentic risk

Overview

The skill provides procurement searches through a disclosed external service, with optional device-based registration that requires consent.

Search terms and company identifiers go to Zhiliaobiaoxun, and queries may consume credits. Configure your own API key to skip device-based registration, or review the device collection notice before agreeing. Protect the saved API key and temporary login links. All bundled instructions were reviewed, but the external service's privacy practices and runtime behavior were not verified.

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 (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code performs host-device fingerprinting rather than procurement information retrieval or analysis. Accessing and hashing a local MAC address is not disclosed in the declared purpose, and no necessary connection to the stated functionality is established. The chunk does not itself transmit the identifier or perform registration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code performs local device fingerprinting rather than procurement-related retrieval or analysis. Neither the description nor the declared permissions disclose this capability, and the chunk shows no connection making it a necessary supporting implementation detail. It does not show transmission of the fingerprint or completion of any registration.

Content

No source excerpt is available for this finding.

The onboarding flow collects and transmits device fingerprints, a capability outside the manifest's procurement-query and analysis purpose (L052–L055).

Medium
Category
Excessive Agency
Confidence
87% confidence
Finding

When no API key is available, the skill instructs the agent to collect platform, architecture, and a MAC-derived hash and submit them to an account-registration endpoint. Device fingerprinting is not a direct requirement of procurement searches or company analysis and is not disclosed in the manifest; requiring user consent mitigates the privacy risk but does not resolve the purpose mismatch.

Content

No source excerpt is available for this finding.

The query workflow sends user-supplied company names and search criteria to a third-party API without a user-facing privacy disclosure.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The company-analysis workflow instructs the agent to submit company identifiers to external Zhiliaobiaoxun endpoints, while other query workflows similarly transmit user-selected search criteria. Although the API endpoints are documented, the skill does not instruct the agent to tell users that their query details will leave the conversation; the consent notice at L052–L054 covers device characteristics for registration only.

Content

No source excerpt is available for this finding.

The registration workflow collects and transmits a hardware-derived device fingerprint, a capability outside the manifest's procurement-search and analytics purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The instructions read a network interface's MAC address, hash it, and submit the resulting identifier with platform and architecture information to an account-registration endpoint. Although the workflow requires user consent and explains trial deduplication, hardware-based device identification is not a direct requirement of procurement queries or ordinary API authentication and is not declared in the manifest.

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**

L393–L395 mandate an upgrade-notification and acknowledgement workflow outside the stated procurement-analysis purpose.

Low
Category
Excessive Agency
Confidence
88% confidence
Finding

The instructions require a per-session account request to check upgrade notices, relay the notice to the user, and submit a POST acknowledgement recording its display. Package-update notification tracking is separate from the procurement searches and analytics declared in the manifest, and the acknowledgement modifies remote notification state rather than retrieving procurement information.

Content

No source excerpt is available for this finding.