Back to skill

Security audit

招投标快捷检索引擎-标找找

Security checks for vulnerabilities and agentic risk

Overview

The search workflow is coherent, but update acknowledgments use your API key without clearly restricting where those requests may go.

Before installing, ask for update acknowledgments to use a fixed provider endpoint with destination validation. You can avoid device fingerprinting by supplying your own API key; otherwise, understand that the MAC hash can recognize the same device over time. Searches send your criteria to the provider and may consume credits. Protect the stored key and generated login links. All bundled files were inspected, but server behavior and data retention 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 (15)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared purpose concerns retrieving and summarizing procurement announcements. The supplied code instead reads local system network information and hashes a MAC address, an undeclared capability with no evident supporting role in that search workflow. The chunk does not show any search, announcement extraction, registration request, or external transmission.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The code generates a device-derived identifier rather than searching procurement announcements. Accessing and hashing a MAC address is not disclosed in the description and is not an evident supporting detail of the declared search function. The snippet does not show transmission of the identifier or actual registration.

Content

No source excerpt is available for this finding.

Credential onboarding collects device fingerprints and creates an account, capabilities not justified by the declared search scope.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

When credentials are absent, L052–L056 direct the agent to collect platform, architecture, and a MAC hash, transmit them to an account-registration endpoint, and persist the returned credential locally. Although the instructions require user consent, device fingerprinting and account creation go beyond using an API key to search tender announcements and are not disclosed in the manifest.

Content

No source excerpt is available for this finding.

The skill exposes corporate due-diligence queries beyond its declared tender-announcement search purpose.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest at L003 describes keyword searches for tender or award announcements, returning project names, amounts, and links. The tool instructions additionally enable corporate registration, shareholder, investment, and corporate-change queries at L074–L084, extended to annual reports, certificates, and credit ratings at L536–L548; these are independent corporate-research capabilities rather than implementation details of announcement search.

Content

No source excerpt is available for this finding.

The skill mandates an authenticated upgrade-notification workflow outside its declared tender-search purpose.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The manifest describes keyword-based tender searches returning concise project names, amounts, and links. L097–L099 instead require a first-use account request to check upgrade notices, verbatim delivery of server-provided messages, and an authenticated POST acknowledging their display. This mandatory notification and remote-state update workflow is not necessary to perform the declared search task and runs without a user requesting account or upgrade management.

Content

No source excerpt is available for this finding.

Automatic company searches lack a privacy disclosure about transmitting user-provided queries to an external service.

Medium
Category
Output Handling
Confidence
84% confidence
Finding

These workflow instructions direct the agent to send the user's company name to search_company and reuse the matched company list in subsequent queries without interrupting the user. Unlike a passive API catalog entry, this selects an actual data-transmission workflow, but does not warn that the user's company research interests will be shared with the external API provider.

Content

No source excerpt is available for this finding.

The skill documents company intelligence and contact-retrieval capabilities beyond quick tender searching.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

The manifest describes keyword-based tender and award searches returning project names, amounts, and links. L167-L303 instead document customer/supplier relationship analysis, retrieval of named project contacts and phone numbers, competitor analysis, and supplier recommendations; these are distinct business-intelligence capabilities rather than necessary search implementation details. The phone-masking and export restrictions at L237 limit exposure but do not justify contact retrieval within the stated scope.

Content

No source excerpt is available for this finding.

Mandatory account-based upgrade checks and display acknowledgments are outside the tender-search purpose.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

L393 requires an authenticated account-balance request at the first use in every session to check for package upgrades, rather than to retrieve tender announcements. L394–L395 additionally require forwarding upgrade notices and submitting authenticated POST acknowledgments of their display. This package-update workflow is not necessary for the manifest's keyword search and concise project-name, amount, and link results.

Content

No source excerpt is available for this finding.

The skill defines paid corporate due-diligence capabilities unrelated to its declared tender-announcement search purpose.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

The manifest limits the skill to keyword searches for tender or award announcements and concise project names, amounts, and links. The reference instead provides actionable requests for corporate registration, shareholders, investments, annual-report financial and contact information, and credit ratings (L31–L111), with successful queries charged credits under L24; these are independent company-research capabilities rather than necessary tender-search implementation details.

Content

No source excerpt is available for this finding.

L334–L336 require an account-level upgrade-notification and acknowledgment workflow unrelated to tender searches.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The workflow requires verbatim publication of server-controlled notice text and an authenticated follow-up based on a response-supplied display_ack. Without explicit message and destination validation, a compromised response could introduce phishing content or redirect an authenticated acknowledgment outside its intended scope. Checking versions is not inherently unsafe, but this extra workflow expands the trust boundary beyond tender searches.

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 account enrollment collects and transmits a persistent hardware-derived identifier beyond the stated tender-search purpose.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

Registration transmits a stable hardware-derived identifier while describing the collected features as having no identifying significance; hashing a MAC address does not eliminate device linkability. The explicit consent requirement and existing-key bypass reduce risk, but the implementation-oriented pseudocode omits the consent gate, making accidental collection without approval plausible.

Content

No source excerpt is available for this finding.

Mandatory account-based upgrade checks and display acknowledgments exceed the declared search workflow.

Low
Category
Excessive Agency
Confidence
97% confidence
Finding

L351 requires an authenticated account-balance request on the first use of each session specifically to check upgrade notices, rather than to retrieve tender results. L352-L353 additionally require relaying notices and posting a display acknowledgment, introducing package-maintenance messaging and remote acknowledgment state into a skill advertised as producing concise search lists. These instructions establish extra behavior, although this reference file alone does not demonstrate that the requests have executed.

Content

No source excerpt is available for this finding.