Back to skill

Security audit

权威采招政策与标讯指南-元博网

Security checks for vulnerabilities and agentic risk

Overview

The procurement tools fit their purpose, but optional signup sends a stable device identifier while understating its privacy implications.

Review the device-identification wording before accepting automatic signup, and ask the provider about retention and deletion. Configure your own ZLBX_API_KEY to skip hardware collection. The documented service is Zhiliaobiaoxun, and data queries can spend account credits. Protect the saved API key and generated login links. All bundled instructions were reviewed; server-side identifier handling 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
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (12)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs hardware-identifier collection and hashing rather than procurement data aggregation or analysis. This is an undeclared capability with no evident supporting role in the stated purpose. The snippet does not show transmission or storage of the resulting hash.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

声明用途是采招信息检索与市场分析,而代码实际访问 /sys/class/net 获取本机网络硬件信息并生成哈希。该行为属于未声明的设备标识能力,与标讯聚合分析没有明确关联,不能视为所述功能的普通实现细节。代码未展示标讯接口调用、市场简报生成或数据外传。

Content

No source excerpt is available for this finding.

The skill collects and transmits device fingerprints for account registration, beyond its stated procurement-analysis purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

L052–L056 instruct the agent to collect platform, architecture, and a MAC-derived hash, submit them to an account-registration endpoint, and persist the resulting credentials. Authentication is expected for the described data APIs, but device fingerprinting and device-linked account creation are additional capabilities not justified by the manifest's market-briefing purpose; the consent requirement limits exposure but does not resolve that scope mismatch.

Content

No source excerpt is available for this finding.

The skill mandates an upgrade-notification and acknowledgment workflow beyond its stated procurement-analysis purpose.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest describes aggregating and analyzing procurement information to produce market briefings. L097–L099 instead instruct the agent to query the account endpoint on first use in every session, relay server-provided upgrade messages verbatim, and submit an authenticated POST acknowledgment that changes the server's notification state. This mandatory package-maintenance workflow is separate from obtaining or analyzing procurement data.

Content

No source excerpt is available for this finding.

The skill exposes individual project-contact lookup beyond its stated market-briefing purpose (L220–L237).

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The manifest describes infrastructure and procurement research using aggregation and analysis to produce market briefings. L220–L237 instead document retrieval of individual project contacts, including complete telephone numbers for paid accounts; this personal-contact discovery capability is not necessary for the stated aggregate-analysis purpose.

Content

No source excerpt is available for this finding.

Mandatory package-update checks and display acknowledgments introduce an unrelated lifecycle-management workflow (L351–L353).

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

L351 requires an authenticated account request at the first use in every session specifically to check upgrade notices, and L352–L353 require relaying those notices and submitting a POST display acknowledgment. This package-update and acknowledgment capability is separate from the aggregation and procurement analysis promised by the manifest, rather than an API operation needed to produce the requested market brief.

Content

No source excerpt is available for this finding.

The reference expands a bid-policy and market-briefing skill into standalone corporate ownership research.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest describes infrastructure and procurement queries and macro-trend briefings based on bidding data. L120–L132 instead prescribe a standalone shareholder-query workflow, explicitly directing the agent to call get_company_shareholders when the user asks only about shareholders, without requiring any procurement or bid-analysis context. This is a separate corporate-research function rather than an implementation detail of the declared briefing purpose.

Content

No source excerpt is available for this finding.

Automatic registration collects and transmits a persistent hardware fingerprint unrelated to procurement-policy and bidding analysis.

Medium
Category
Excessive Agency
Confidence
92% confidence
Finding

The registration flow transmits a deterministic, unsalted MAC-address hash, which remains a persistent pseudonymous device identifier rather than anonymous data; SHA-256 does not prevent correlation or guessing candidate MAC addresses. Explicit prior consent, a configured-key bypass, and the stated trial-deduplication purpose reduce the risk, but categorical claims that these features have no identifying significance understate the privacy implications and weaken informed consent.

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 mandates an upgrade-notification and acknowledgment workflow unrelated to procurement market analysis.

Low
Category
Excessive Agency
Confidence
88% confidence
Finding

L393–L395 instruct the agent to query an account endpoint each session for upgrade notices, relay those notices, and submit an authenticated POST acknowledgment after display. This introduces package-maintenance messaging and remote acknowledgment behavior beyond the manifest's stated purpose of analyzing procurement notices and producing market briefs; it is not necessary to retrieve or aggregate procurement data.

Content

No source excerpt is available for this finding.