Back to skill

Security audit

地方标讯极速检索-比地招标

Security checks for vulnerabilities and agentic risk

Overview

This procurement-search skill needs review because its signup notice understates device identification and its mandatory update workflow lacks clear safeguards.

Prefer a manually obtained API key to avoid hardware-based registration. Searches send keywords and filters to Zhiliaobiaoxun, and successful data queries can consume credits. Protect the saved key and automatic-login links, and independently check update links before acting on notices. Keep your requested geography unchanged unless you approve broader results. The package does not establish provider-side identifier retention, deletion practices or live acknowledgment validation; these remain uncertainties.

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 (18)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code performs local device identification rather than regional procurement search. It contains no geographic filtering, announcement retrieval, chronological sorting, or budget extraction. Reading and hashing a MAC address is an undeclared capability with no evident supporting role in the stated purpose. The snippet does not show transmission 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

该代码未执行地域过滤、采购公告查询、时间排序或预算提取,而是处理本机网络硬件标识。此行为不属于所声明检索功能的明显支持性实现细节,因此存在描述与行为不符。代码中未显示对所得标识的外传。

Content

No source excerpt is available for this finding.

Missing credentials trigger device-fingerprint collection and remote account creation outside the stated search purpose.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The fallback workflow instructs the agent to collect platform, architecture, and a MAC-derived hash, submit an automatic-registration request, and persist the resulting credential locally. Although L052-L054 require consent, device identification and account provisioning are distinct capabilities not declared by the regional-search manifest or necessary merely to submit an authenticated tender search.

Content

No source excerpt is available for this finding.

The skill exposes corporate intelligence capabilities beyond localized procurement-announcement search.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest describes returning geographically filtered procurement notices and budget amounts, but these instructions expose shareholder records, corporate investments, project contacts, competitor identification, and market analytics. The executable API examples at L342-L377 and additional corporate-record tools at L536-L548 make these operational capabilities rather than incidental background documentation, including queries that consume paid credits.

Content

No source excerpt is available for this finding.

The search workflow lacks a privacy notice that user-provided query details are sent to an external service.

Medium
Category
Output Handling
Confidence
87% confidence
Finding

The query workflow directs searches through the externally hosted Zhiliaobiaoxun API, transmitting user-derived keywords, company names, locations, and other filters. Although the API endpoints and request formats are documented, the skill does not warn users that their search details will leave the agent environment; the device-data consent requirement applies only to automatic registration.

Content

No source excerpt is available for this finding.

The no-results workflow permits relaxing geography despite the manifest requiring strict regional filtering.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The QUERY_EMPTY guidance permits suggesting broader geography despite the manifest's strict regional-filter requirement. Conflicting instructions can cause an agent to return out-of-region procurement opportunities as responsive results, creating a limited integrity risk rather than a credential or system-compromise risk.

Content

No source excerpt is available for this finding.

The skill adds account-finance and usage-history queries unrelated to regional tender searches.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest limits the skill to geographically filtered procurement announcements and their budget amounts. These instructions instead enable querying private account balances, cumulative recharge and spending totals, and daily usage history, including estimating how long the user's balance will last; these are separate account-management capabilities, not tender-search implementation details.

Content

No source excerpt is available for this finding.

Every session is required to perform an undeclared upgrade-notification and acknowledgement workflow.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The instructions require an account request before the first skill use in each session to check upgrade notices, forwarding the returned message verbatim and then submitting a POST display acknowledgement. This adds unsolicited user messaging and a remote state-changing operation to a skill described only as a regional tender search tool.

Content

No source excerpt is available for this finding.

The skill exposes company-intelligence capabilities beyond geographically filtered tender search.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest limits the skill to retrieving local procurement announcements and budget amounts with strict geographic filtering. These API instructions instead cover corporate profiles, upstream/downstream relationship analysis, company contact retrieval (including full telephone numbers for paid accounts at L237), competitor analysis, and supplier recommendations—independent intelligence capabilities not required for the declared search task.

Content

No source excerpt is available for this finding.

Mandatory account-based upgrade checks and display acknowledgments are unrelated to local tender retrieval.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The instructions require an authenticated account-balance request at first use to obtain upgrade notices, followed by a POST acknowledgment after displaying a notice. This adds an account-management and remote acknowledgment workflow to a skill whose declared purpose is retrieving geographically filtered procurement announcements; neither operation is necessary to perform that search.

Content

No source excerpt is available for this finding.

The skill includes market-analysis capabilities beyond its stated local procurement-search purpose.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

L013–L017 document purchaser, supplier, and brand rankings, multidimensional aggregation, and historical product-price analysis. Their detailed descriptions explicitly target customer acquisition, competitor analysis, and sourcing (L023, L079, L189, L290), rather than returning geographically filtered procurement notices and budgets in reverse chronological order.

Content

No source excerpt is available for this finding.

Mandatory account-based upgrade checks and display acknowledgments are unrelated to local tender searches.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

L393 requires an authenticated account-balance request on the first use of every session specifically to check upgrade reminders. L394–L395 additionally require relaying update messages and submitting an authenticated POST acknowledgment, introducing account and update-notification operations that are not needed to retrieve regional procurement notices.

Content

No source excerpt is available for this finding.

The skill exposes paid corporate due-diligence capabilities unrelated to geographically filtered tender searches.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The manifest limits the skill to regional procurement announcements and budgets, but L031–L111 provide callable instructions for corporate registration, shareholders, investments, corporate changes, annual reports, qualifications, and credit ratings. These are standalone company-investigation capabilities rather than necessary implementation details of local tender retrieval, and L024 specifies that successful queries consume user credits.

Content

No source excerpt is available for this finding.

The skill collects and transmits a persistent hardware-derived identifier for account registration, beyond its stated regional tender-search purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The registration flow transmits a deterministic, unsalted MAC-address hash that remains a persistent device identifier; hashing does not make it anonymous or prevent correlation. Although the skill requires consent and offers manual registration, its categorical assurances that no identifying information is collected understate this privacy risk. Device-based trial deduplication explains the collection but is separate from the regional tender-search function.

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 account-based upgrade-notification and acknowledgment workflow unrelated to regional tender search.

Low
Category
Excessive Agency
Confidence
95% confidence
Finding

The workflow requires relaying a server-controlled message verbatim into the user conversation without defining validation or content restrictions. This creates an output-integrity risk: a compromised notification service could insert misleading upgrade instructions or phishing links into an otherwise legitimate tender-search interaction. The same-provider endpoint and display-before-acknowledgment requirement limit the concern; the text does not authorize executing instructions contained in the notice.

Content

No source excerpt is available for this finding.