Back to skill

Security audit

招中标数据智能体-AI驱动的标讯分析Agent

Security checks for vulnerabilities and agentic risk

Overview

The procurement queries fit the skill’s purpose, but trial signup understates the privacy impact of sending a persistent device identifier.

Review the device-identification disclosure before accepting automatic signup. Configuring an existing ZLBX_API_KEY skips hardware enrollment. Queries send your search targets to Zhiliaobiaoxun and may consume account credits; update notices also generate display receipts. Protect the saved API key and avoid confidential query terms unless you accept the provider’s handling of them. The artifacts do not establish server-side retention or deletion practices.

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 (11)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Enrollment transmits a deterministic MAC-derived identifier that can link a device to an account, while describing the collected characteristics as having no identity significance. Explicit consent and a manual-registration alternative reduce the risk, but hashing does not make this persistent identifier anonymous; the issue is misleading privacy characterization rather than merely functionality outside the description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The registration workflow creates a stable hardware-derived identifier and sends it to the provider, creating a device-linkability risk not conveyed accurately by the non-identifying-data claim. Requiring consent and allowing existing credentials mitigate this concern, but do not make the identifier anonymous.

Content

No source excerpt is available for this finding.

The credential-enrollment workflow collects and transmits device fingerprints beyond the skill's stated bidding-data query and analysis purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The enrollment branch transmits a stable MAC hash alongside platform information, permitting device-account linkage. The opt-in requirement limits exposure, but the referenced disclosure misleadingly characterizes these features as non-identifying, weakening informed consent.

Content

No source excerpt is available for this finding.

The query workflow sends user-supplied search terms and company names to an external service without a user-facing privacy disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The company-analysis workflow instructs the agent to submit company names to multiple Zhiliaobiaoxun API endpoints. Although the file documents these endpoints and requires consent for device-feature collection during registration, it does not instruct the agent to disclose that ordinary queries also transmit user-provided information to this third-party service.

Content

No source excerpt is available for this finding.

The automatic company-analysis workflow sends user-supplied company names to an external API without a user-facing privacy disclosure.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

This workflow explicitly instructs the agent to call search_company with the user's company name and use the matched names in subsequent queries, rather than merely listing available tools. It does not instruct the agent to disclose that these search targets will be transmitted to the external Zhiliaobiaoxun service, potentially revealing the user's business research interests.

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**

Automatic registration collects and transmits a persistent hardware fingerprint beyond the skill's bidding-data analytics purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The workflow sends an unsalted, normalized MAC hash that is stable across installations using the same hardware address, yet its consent notice claims no personal or identity-bearing information is collected. Hashing limits direct exposure of the MAC but preserves linkability and permits candidate-MAC comparison; explicit opt-in mitigates rather than eliminates the misleading-disclosure risk.

Content

No source excerpt is available for this finding.

The skill mandates a package-update notification and display-tracking workflow outside its stated bidding-data query and analysis purpose.

Low
Category
Excessive Agency
Confidence
86% confidence
Finding

L097 requires an account request at the start of each session specifically to check for package-update notices. L098–L099 then require relaying the notice and submitting an authenticated POST acknowledging its display, adding remote notification-state management unrelated to the requested bidding-data analysis.

Content

No source excerpt is available for this finding.

The skill reports that the user was shown an update notice without disclosing this reporting to the user.

Low
Category
Output Handling
Confidence
88% confidence
Finding

L336 instructs the agent to submit an authenticated POST acknowledgment containing the notice ID and displayed: true after showing an update notice. This transmits user-interaction information to the API provider, but the file provides no user-facing privacy warning and explicitly says not to show the technical acknowledgment.

Content

No source excerpt is available for this finding.