Back to skill

Security audit

tender-search

Security checks for vulnerabilities and agentic risk

Overview

The procurement features fit their purpose, but conflicting registration instructions and understated device identification warrant review.

Prefer a manually configured API key and require account recovery when credentials fail. Ask the publisher to reconcile registration eligibility and include consent in the example flow. Treat the MAC hash as a persistent device identifier, keep keys and login links private, and avoid confidential search terms. All bundled files were reviewed; server-side retention and deletion practices remain unverified.

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 (10)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared purpose concerns procurement-data queries and analysis, not collection or hashing of local hardware identifiers. This chunk performs an undeclared fingerprinting capability with no demonstrated connection to those tasks. It does not itself transmit the hash or perform registration.

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 local device fingerprinting rather than tender or procurement data analysis. Accessing and hashing a MAC address is not disclosed in the description, and the chunk establishes no connection to the declared functionality. It does not show transmission, storage, or registration of the resulting hash.

Content

No source excerpt is available for this finding.

The search workflow sends user-provided query details to an external service without a privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow directs the agent to execute searches using user-provided keywords and filters through the external API configured at L019–L034. Although device-feature collection during registration has an explicit consent step, routine searches have no comparable disclosure that company names, procurement interests, and other potentially confidential query details leave the user's environment.

Content

No source excerpt is available for this finding.

Authentication-error handling contradicts the explicit restriction on automatic registration.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The authentication-error table directs automatic registration even when an existing key is invalid, expired, or disabled, contradicting the explicit restriction to users without configured credentials. An agent following this branch could initiate unnecessary device collection or replace the intended account credentials; the conflicting instructions make the account and consent boundaries unreliable.

Content

No source excerpt is available for this finding.

The skill mandates a package-update notification and acknowledgement workflow unrelated to tender search or analysis.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

These instructions require an account request on the first use of each session to check for upgrade notices, relay server-provided notices, and submit an authenticated POST acknowledging their display. Package-update tracking and remote acknowledgement are separate capabilities from the procurement search and analytics described in the manifest, and they run without a user requesting package maintenance.

Content

No source excerpt is available for this finding.

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

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow instructs 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 directs actual transmission of user-supplied query data to the external API, but provides no instruction to disclose that transmission to the user.

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 collects and transmits a persistent hardware fingerprint for trial-account deduplication, beyond its stated procurement-search and analytics purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

Although the prose requires consent, the supplied get_api_key pseudocode proceeds directly from missing credentials to device-feature collection and external registration without a consent check. An implementation following that example could transmit a persistent MAC-derived identifier without approval, while the description of these features as having no identity meaning understates their linkability.

Content

No source excerpt is available for this finding.