Back to skill

Security audit

招中标信息&招标雷达

Security checks for vulnerabilities and agentic risk

Overview

The search features fit their purpose, but the update workflow leaves unclear where your account key may be sent.

Before installing, have the acknowledgment workflow restricted to a fixed provider endpoint so response content cannot redirect your account key. Existing credentials avoid device fingerprinting. Search terms leave your environment, queries may consume credits, and saved keys and generated login links should remain private. All seven bundled files were reviewed; the live acknowledgment format remains 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 (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local device fingerprinting rather than procurement-data querying or analysis. Reading and hashing a MAC address is an undeclared capability with no demonstrated supporting role in the stated purpose. The snippet does not show transmission or storage of the resulting identifier.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code computes a hash derived from the machine's MAC address rather than querying or analyzing procurement data. Device identification is not disclosed in the description, and the snippet establishes no necessary supporting relationship to the declared functionality. No transmission or storage of the resulting hash is shown.

Content

No source excerpt is available for this finding.

The query workflow lacks a user-facing notice that search criteria are sent to an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The execution rules instruct the agent to run searches using user-supplied criteria, which the API instructions send to zhiliaobiaoxun.com. Although device-feature collection during registration requires consent, ordinary queries have no equivalent disclosure that company names, product interests, and other potentially sensitive business criteria leave the user's environment.

Content

No source excerpt is available for this finding.

L097–L099 introduce a mandatory package-update notification and remote display-acknowledgment workflow unrelated to tender-data analysis.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

An update-notification workflow is not inherently unsafe, but this one directs authenticated acknowledgment requests according to a response-provided display_ack without specifying a fixed endpoint or destination validation. If that field can select an external destination, a malicious or compromised response could cause the agent to forward its API credential outside the intended service.

Content

No source excerpt is available for this finding.

The skill mandates an account-based upgrade-notification and acknowledgment workflow unrelated to tender-data analysis.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The ancillary update check is legitimate in principle, but the acknowledgment instructions reuse credentials while leaving the response-provided display_ack destination insufficiently constrained. If interpreted as a request destination, an attacker-controlled value could redirect an authenticated request and expose the API key.

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 skill collects and transmits a hardware-derived device fingerprint for trial-account registration, beyond its stated tender-data analytics purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The registration branch collects platform, CPU architecture, and a MAC-address hash and submits them to an account-registration endpoint. Unlike API authentication or tender-data requests, hardware fingerprinting for trial deduplication is not a direct requirement of the manifest's search, pricing, forecasting, or supplier-recommendation functions; the documented consent gate mitigates privacy risk but does not bring this capability within the declared purpose.

Content

No source excerpt is available for this finding.