Back to skill

Security audit

海量标讯智搜助手-标800

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because its signup privacy wording understates device identification and its update acknowledgments leave credential handling insufficiently scoped.

Before installing, request accurate device-identifier disclosure and a fixed, validated same-origin acknowledgment endpoint. Configure your own API key to skip device-based registration. Searches send selected criteria to Zhiliaobiaoxun, and paid lookups consume credits. Protect the saved key and temporary login links. Live API behavior was not tested.

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 (15)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local device identification rather than constructing or filtering tender searches. Accessing and hashing a MAC address is not disclosed in the description and has no evident supporting role in the declared search functionality. The snippet does not show transmission of the identifier or any actual registration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The code derives a fingerprint from a local network interface rather than constructing or executing tender-information searches. Hardware-identifier access and fingerprint generation are not disclosed by the declared search purpose and have no demonstrated supporting role in it. The supplied chunk does not transmit the identifier or perform registration.

Content

No source excerpt is available for this finding.

L052-L056 direct device-fingerprint collection, remote account registration, and persistent credential storage beyond tender searching.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest describes combining search conditions and returning filtered tender information, but these instructions additionally collect platform, architecture, and a MAC-derived hash, transmit them for account registration, and write credentials to disk. Explicit user consent reduces the privacy concern, but device fingerprinting and account provisioning remain separate capabilities not declared in the manifest's search purpose.

Content

No source excerpt is available for this finding.

L074-L080 and L536-L548 expand a complex tender-search skill into corporate registry and credit-information retrieval.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest promises precise tender searches using keywords, exclusions, and amount ranges, whereas the operational tool list also supports shareholder records, investments, and corporate registration changes. L536-L548 further authorize annual-report, certification, and historical credit-rating queries, which are independent corporate-information services rather than implementation details of tender filtering.

Content

No source excerpt is available for this finding.

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

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The workflow instructs the agent to execute searches using user-supplied criteria through the external Zhiliaobiaoxun API, but does not require informing the user that those criteria leave the conversation. Company names, keywords, exclusions, and budget ranges can reveal commercially sensitive interests; the device-data consent for automatic registration does not cover search-query disclosure.

Content

No source excerpt is available for this finding.

Automatic company lookup transmits user-supplied company names to an external API without a privacy notice.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow instructs the agent to send the company name from the user's request to search_company and use the matched names in subsequent queries without interrupting the user. Unlike a passive API catalog entry, this is an explicit execution instruction, but the file provides no user-facing disclosure that these query details are sent to the external 知了标讯 service.

Content

No source excerpt is available for this finding.

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

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

L351 requires an authenticated account-balance request on first use of each session specifically to check upgrade notices, while L352–L353 require relaying those notices and posting a display acknowledgment. This adds account access and remote notification-state updates beyond the manifest's purpose of combining tender-search criteria and filtering results; the instructions do not establish these operations as prerequisites for searching.

Content

No source excerpt is available for this finding.

The skill mandates account-level upgrade checks and display acknowledgements unrelated to tender search (L393–L395).

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The instructions require an authenticated account-balance request at the first use in each session to check upgrade notices, followed by a POST acknowledgement after displaying a notice. This adds account access and server-side notification-state changes for package lifecycle management, neither of which is necessary for the manifest's stated purpose of combining and filtering tender-search conditions.

Content

No source excerpt is available for this finding.

The skill exposes paid corporate due-diligence operations beyond its declared tender-search purpose.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest describes precise tender searches using keywords, exclusions, and amount ranges. Instead, L024–L111 define paid corporate-registry operations covering shareholders, investments, corporate changes, annual financial reports, and credit ratings; these are standalone company-research capabilities, not implementation details of tender filtering. Although the reference restricts calls to user-requested capabilities, it still enables a materially broader, billable workflow than the manifest declares.

Content

No source excerpt is available for this finding.

L334–L336 require authenticated account access and update-display acknowledgments unrelated to tender searching.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An authenticated balance check is not inherently unsafe, but this workflow additionally requires verbatim display of remotely supplied messages and an authenticated POST guided by response-supplied display_ack data. Without explicit destination and content validation, a compromised response could inject deceptive user-facing instructions or potentially redirect a credential-bearing acknowledgment; these extra operations are unnecessary for tender searching.

Content

No source excerpt is available for this finding.

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

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The registration flow sends a stable, unsalted hardware-derived identifier to an external service while describing it as having no identifying meaning and collecting no personal information. Hashing a MAC address does not make it anonymous: the result remains linkable and may permit candidate enumeration, so the consent wording understates the privacy implications. Explicit consent and the existing-key bypass reduce exposure but do not correct that misleading assurance.

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**

L097–L099 require an unrelated package-update workflow in a skill described only as precise tender searching.

Low
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest describes combining tender-search conditions and filtering results, but these instructions require an authenticated account request at first use in each session to check package-update notices. They also require relaying update messages and conditionally submitting a POST acknowledgment that changes server-side notification state; neither operation is needed to execute a tender search.

Content

No source excerpt is available for this finding.