Back to skill

Security audit

全国采招大数据中心-采招网

Security checks for vulnerabilities and agentic risk

Overview

The skill supports procurement research with disclosed account access and optional device-based registration that requires consent.

Search terms and company names are sent to 知了标讯, and queries can consume account credits. Configure an existing API key to skip device-based enrollment; otherwise, consider the MAC hash a persistent identifier before consenting. Protect the saved key and generated login links. This review covers the bundled instructions, not the service's retention practices or live security controls.

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
99% confidence
Finding

The supplied code performs local device fingerprinting rather than procurement-data retrieval or enterprise profiling. Reading and hashing a MAC address is not an evident supporting detail of the declared functionality and accesses an inconsistent resource. The chunk does not show external transmission or use of the resulting hash.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code performs local device fingerprinting rather than procurement searches or enterprise-profile analysis. Nothing in the declared purpose explains why a MAC-derived identifier is needed, so this is an undeclared capability rather than an evident supporting implementation detail. The snippet does not show transmission or storage of the identifier.

Content

No source excerpt is available for this finding.

The registration workflow collects and transmits device fingerprints, a capability unrelated to procurement searches or enterprise analysis.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

L054–L055 instruct the agent to collect platform, architecture, and a MAC-derived hash and submit those device characteristics for account registration. Device fingerprinting is not required for the manifest's procurement retrieval and enterprise-profiling purpose, which already supports authentication with a supplied API key. The explicit consent requirement mitigates the privacy risk but does not make device identification part of the stated analytical scope.

Content

No source excerpt is available for this finding.

The skill adds a mandatory package-update notification and acknowledgment workflow unrelated to procurement search or enterprise profiling.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

L097 requires an authenticated account request at the first use of every session specifically to check upgrade notices, rather than to fulfill a procurement query. L098–L099 then direct the agent to relay the server-provided message verbatim and submit an authenticated POST acknowledging its display, introducing remote notification handling and state modification outside the manifest's stated search and enterprise-analysis purpose.

Content

No source excerpt is available for this finding.

Automatic company searches send user-provided company names to an external API without a privacy disclosure.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The file instructs the agent to call search_company with a company name supplied by the user, rather than merely listing an available tool. Although it documents the request parameters, it does not warn users that their query information will be transmitted to the external 知了标讯 service.

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 enrollment collects and transmits a hardware-derived device fingerprint beyond the stated procurement-analysis purpose.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The manifest describes procurement searches and enterprise profiling, but the automatic-registration branch collects operating-system, architecture, and MAC-derived fingerprint data and sends it to an account-creation endpoint. The MAC collection commands at L065–L099 and registration payload at L109–L125 establish a device-identification capability for trial-account deduplication, not procurement analysis. The consent requirement limits when this happens but does not make hardware fingerprinting a declared or necessary part of the skill's purpose.

Content

No source excerpt is available for this finding.

The skill mandates an authenticated upgrade-notice and acknowledgment workflow unrelated to procurement searches or company analysis.

Low
Category
Excessive Agency
Confidence
88% confidence
Finding

L351 requires an account-balance request at the start of each session specifically to check upgrade reminders, and L353 directs a subsequent authenticated POST to record that a notice was displayed. This package-update workflow, including modification of server-side acknowledgment state, is separate from the manifest's stated procurement-data retrieval and enterprise-profiling purpose.

Content

No source excerpt is available for this finding.

L393–L395 require a session-level upgrade-notification and acknowledgement workflow unrelated to procurement research.

Low
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest describes procurement searches and comprehensive company analysis, but these instructions additionally require checking for package upgrades, relaying server-provided notices, and submitting a POST receipt confirming their display. This user-facing notification and remote acknowledgement capability is separate from authenticating or retrieving procurement data and is not justified by the stated purpose.

Content

No source excerpt is available for this finding.