Back to skill

Security audit

招投标AI数据分析平台

Security checks for vulnerabilities and agentic risk

Overview

Review is recommended because optional signup sends a persistent device fingerprint while describing it as non-identifying.

Configure an existing API key or choose manual registration to avoid hardware fingerprint collection. If accepting automatic signup, understand that hashing still permits device recognition; the artifacts do not explain retention or deletion. Search terms go to Zhiliaobiaoxun, and paid lookups consume credits. Protect the stored key and temporary login links. Server-side privacy practices were 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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (13)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code performs device identification rather than procurement-data analysis. The declared purpose does not disclose device fingerprinting or establish why accessing a MAC address is necessary as a supporting implementation detail. The chunk 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
97% confidence
Finding

The supplied code performs device identification rather than procurement-data analysis. Reading and hashing a local MAC address is an undeclared capability with no demonstrated connection to the described analytical functions. Although the file name suggests registration support, the supplied material does not establish why hardware fingerprinting is needed. The code does not show transmission of the resulting hash.

Content

No source excerpt is available for this finding.

The registration workflow collects and transmits device fingerprints beyond the stated bidding-analysis purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

L052–L055 instruct the agent to collect platform, architecture, and a MAC-derived hash and submit those device characteristics during automatic registration. Device fingerprinting is not a direct requirement of procurement-market analysis and is not declared in the manifest description. The required user consent mitigates unauthorized collection but does not make this capability necessary for the stated analytical purpose.

Content

No source excerpt is available for this finding.

The query workflow sends user-supplied search criteria to an external service without a privacy disclosure.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The company-analysis workflow instructs the agent to transmit company names and related query criteria to the Zhiliaobiaoxun API, but does not tell users that their supplied information will leave the conversation for external processing. The consent requirement at L052–L055 covers device-feature collection during registration, not ordinary query submissions made with existing credentials.

Content

No source excerpt is available for this finding.

The skill adds a mandatory package-update notification and remote display-acknowledgment workflow unrelated to bidding analytics.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The manifest describes bidding-market analysis, aggregation, and forecasting, but these instructions require an account request at the first use in every session specifically to check for upgrade notices. They also direct the agent to relay a server-provided message and submit an authenticated POST confirming its display, introducing package-maintenance and acknowledgment reporting beyond the stated analytical purpose.

Content

No source excerpt is available for this finding.

Automatic company searches transmit user-provided company names to an external service without a privacy disclosure.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The file instructs the agent to call search_company when a user supplies a company abbreviation, and lines 75–82 extend this into automatic downstream queries without user confirmation. This is an executable workflow rather than a tool listing, but it does not warn that the user's company names and research targets will be sent to the external 知了标讯 service.

Content

No source excerpt is available for this finding.

The skill exposes standalone corporate ownership and annual-report lookups beyond its declared tender-market analytics purpose (L048–L079).

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes procurement-market aggregation, market-share analysis, historical winning prices, and candidate prediction. These instructions additionally enable paid queries for shareholders, branches, investments, registration changes, and annual reports containing financial statements and telephone/email details; this general corporate-investigation capability is not declared or restricted to supporting a tender-analysis task.

Content

No source excerpt is available for this finding.

The skill collects and transmits a hardware-derived device fingerprint for trial-account deduplication, beyond its stated bidding-analysis purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The workflow transmits a stable, unsalted hash of a hardware address, enabling device recognition and potential linkage to account activity; hashing does not make this identifier anonymous, and constrained MAC-address ranges can permit guessing. Explicit opt-in and the configured-key bypass reduce the risk, but describing the fingerprint as having no identifying significance understates its privacy implications and weakens 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 unrelated upgrade-notification and display-tracking workflow.

Low
Category
Excessive Agency
Confidence
87% confidence
Finding

L351–L353 require a per-session account request to check upgrade notices, verbatim delivery of server-provided notices, and an authenticated POST acknowledging their display. This adds user-facing package-maintenance messaging and remote notification-state updates beyond the manifest's bidding analytics purpose; these operations are not necessary to perform market analysis or supplier prediction.

Content

No source excerpt is available for this finding.

L334–L336 mandate an upgrade-notification and acknowledgement workflow unrelated to bidding analysis.

Low
Category
Excessive Agency
Confidence
86% confidence
Finding

The instructions require checking for package upgrades, relaying the server-provided notice to the user, and submitting a POST acknowledgement recording its display. This package-maintenance workflow is not declared in the manifest and is not needed to retrieve or analyze procurement data; unlike ordinary API authentication, it introduces a separate notification-delivery capability and remote state change.

Content

No source excerpt is available for this finding.