Back to skill

Security audit

招投标商机监控雷达-标标达

Security checks for vulnerabilities and agentic risk

Overview

The procurement tools are largely coherent, but trial registration understates the privacy implications of sending a persistent device identifier.

Review the registration privacy wording before opting in. Configure an existing ZLBX_API_KEY to skip device collection entirely. If you choose automatic registration, understand that a persistent device identifier goes to ai.zhiliaobiaoxun.com; its retention is unspecified. Protect the saved API key and login links, and expect data queries to consume service credits. Artifact coverage gaps were resolved, but server-side data handling was 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
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied command generates a device fingerprint rather than performing procurement or bidding analysis. The declared purpose does not disclose device identification, and the snippet provides no connection showing that this hardware access is a necessary supporting implementation detail. No transmission, storage, or actual registration is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code generates a device identifier rather than performing procurement or bidding analysis. Reading and hashing a local MAC address is an undeclared capability with no demonstrated connection to the described business functions. The chunk does not show transmission, storage, or actual account registration, so those behaviors cannot be inferred.

Content

No source excerpt is available for this finding.

The manifest promises bidding-risk assessment, but the operational instructions restrict the skill to data queries and redirect bidding decisions elsewhere.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

L003 advertises one-click project feasibility assessment, detection of predetermined or restrictive tenders, and supplier-relationship analysis, explicitly requiring activation for those requests. In contrast, L508 directs the agent not to handle project-specific bidding decisions or deep company investigations and states that this skill focuses on data queries; L499–L501 assign the advertised analysis to other skills. This is a substantive mismatch between the advertised service and the instructed behavior, rather than an implementation detail.

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**

The automatic-registration flow collects and transmits a persistent device fingerprint, a capability outside the stated tender-analysis purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

Optional registration and the explicit consent gate reduce the risk, but the consent text incorrectly describes a stable MAC-derived fingerprint as having no identifying meaning and collecting no personal information. An unsalted MAC hash remains a linkable device identifier, so users may authorize transmission without understanding its tracking implications; hashing is not anonymization.

Content

No source excerpt is available for this finding.