Back to skill

Security audit

专利数据质检

Security checks for vulnerabilities and agentic risk

Overview

The skill performs local patent-table checks and clearly separates optional online lookups, though costs and cached credentials need care.

Local checks need no account or network access. Before requesting online enrichment, approve which information may be sent and set a spending limit. Keep the PatentMax cache private, prefer temporary use of your own key, and follow SKILL.md's storage controls over older FAQ suggestions.

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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (48)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
100% confidence
Finding

The supplied shell command performs no patent-data validation. Accessing and printing an API-key prefix is unrelated to the declared CSV inspection purpose and constitutes undeclared credential access and partial disclosure. No network transmission is shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
100% confidence
Finding

The code has a materially different primary purpose from the description. Its CLI exposes search, brief, similar, citation, figure, stats, and company commands backed by external API endpoints. It has no supplied-CSV inspection workflow for missing columns, empty cells, date precision, chronological inconsistencies, or duplicate source identifiers. Publication-number validation and candidate deduplication support API retrieval and search tracking; they do not implement the declared table audit. External API access, credential acquisition, and potentially chargeable retrieval are substantive undeclared capabilities, not supporting details of local CSV validation. The code can export a candidate ledger to CSV, but exporting search results is not inspecting a supplied CSV.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The command does not read or validate supplied patent CSV data. Instead, it performs an authenticated search against api.ip930.com using a fixed topic query. This is a materially different capability and resource access, not a supporting implementation detail of the declared table-quality checks. The command does not appear to modify source records.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command performs a hard-coded patent search against an external service. It neither reads supplied CSV data nor implements the declared quality checks. External patent retrieval is a materially different capability, not a supporting implementation detail of the described table inspection. The command does not appear to modify source records, but that alone does not make its behavior consistent with the declared purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command performs a remote patent-content lookup and does not read or validate CSV data. External claims retrieval is not a supporting implementation detail for any of the declared checks shown here and accesses a resource outside the described table-inspection workflow. The command does not modify source records, but that alignment does not resolve the capability mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The command performs patent discovery, not inspection of supplied patent data. It does not reference an input CSV or invoke any declared validation checks. The client's underlying implementation is not shown, so network access cannot be confirmed, but the explicit search operation is already outside the declared purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared scope is quality checking user-supplied CSV data, but the chunk documents patent-information search and retrieval commands instead. These capabilities are materially different from checking table completeness, formatting, dates, and duplicates, and are not necessary implementation details of those checks. The client implementation is not supplied, so its precise network behavior and billing cannot be verified; nevertheless, the capabilities explicitly exposed by these commands establish a clear mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands perform patent discovery and similarity retrieval rather than quality checks on supplied CSV records. No command shown reads a CSV or checks missing data, identifier formats, dates, or duplicates. Although both concern patents, the demonstrated capabilities have a materially different purpose and are not supporting implementation details of the declared table-validation workflow.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied command does not read or validate a CSV, check missing values, inspect dates, or detect duplicates. It instead retrieves information about a fixed patent from an external service. No connection to the declared table-quality checks is shown, so this is an undeclared capability rather than a demonstrated supporting implementation detail. The command does not itself establish that database authenticity verification occurs.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied command performs a remote lookup for a fixed patent number; it does not read supplied CSV data or perform any declared data-quality checks. Although a lookup could support some workflows, no connection to the declared checks is shown here, and external patent-data retrieval is not disclosed. The command does not itself establish that authenticity verification occurs or that source records are modified.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The command executes a search operation; it does not supply or inspect a CSV, check missing values, validate identifiers or dates, or detect duplicates. Patent retrieval is a materially different capability from the declared table-quality inspection. The chunk does not establish the client's underlying data source or whether it accesses a network.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The visible command requests a brief for a particular patent and performs none of the declared CSV validation checks. This is materially different from the stated data-quality purpose. The client implementation is not supplied, so network access and the brief’s exact contents cannot be determined.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The command invokes a search workflow, not the declared CSV quality checks. It does not reference supplied CSV data or perform any visible missing-field, format, date, or duplicate checks. Searching for patents by a specific applicant is an undeclared capability and is not shown to support the described inspection workflow. The client implementation is not supplied, so its precise data source and network behavior cannot be established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The command performs a patent similarity lookup rather than any of the declared table-quality checks. This is a materially different capability, not a supporting implementation detail evident from the supplied chunk. The client implementation is not shown, so its network access and other side effects cannot be established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The commands perform patent research and retrieval operations rather than the declared CSV quality checks. The supplied chunk shows no missing-field, format, date, chronology, or duplicate validation. Although the client's implementation and data source are not shown, these explicit retrieval and search operations are outside the declared purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The shown commands implement patent-candidate curation rather than the declared data-quality checks. In particular, the mark commands change candidate selection metadata, an undeclared write capability. The snippet shows no checks for missing fields, publication-number formats, date precision, chronology, or duplicate identifiers. The client's underlying storage and network behavior are not visible, so no additional access claims can be established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied commands manage a client's search budget and usage rather than inspect supplied CSV records. These operations are not declared and have no demonstrated supporting role in the described local data-quality checks. The chunk does not establish that searches occur, charges are incurred, or external services are contacted, because the client implementation is not supplied.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied commands demonstrate patent-content retrieval and exploration, not checks for missing fields, malformed publication numbers, dates, or duplicates. Although reading patent identifiers and dates could support quality inspection, the claims and description-search operations serve a materially different function. The chunk does not demonstrate source modification, external database verification, or network access.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The command checks https://api.ip930.com/api/v1/health rather than performing the declared patent-table checks. Although a health check can be a supporting implementation detail, this chunk establishes no dependency between that service and the stated inspection workflow. It does not upload patent records, modify source data, or establish database authenticity.

Content

No source excerpt is available for this finding.

Code capability 'network' detected in scripts/patentmax_client.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The script performs real authenticated requests to a third-party service, although the declared skill concerns inspecting supplied CSV records and does not describe remote searches or paid analytics. Invoking those commands can disclose query content and consume an existing account balance without a task-appropriate authorization step; the exact manifest-level permission violation cannot be independently verified.

Content

No source excerpt is available for this finding.

Code capability 'env' detected in scripts/patentmax_client.py but not covered by declared permissions.

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not covered by the declared 'allowed-tools' preapproval. Review the host runtime's actual enforcement policy; runtime permissions are unknown unless separately verified, and this declaration alone does not establish a permission ceiling.

Content

No source excerpt is available for this finding.

The supplied-CSV quality checker instead implements a paid remote patent-search and analytics client.

High
Category
Excessive Agency
Confidence
100% confidence
Finding

The implementation provides remote patent search, retrieval, and analytics rather than the advertised supplied-CSV quality checks. This scope mismatch can cause an agent to send user-derived search information externally and incur charges while appearing to perform a local inspection, and no monetary limit is enforced by default.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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 · SKILL.md (reported line 68)May include surrounding context.

md
### 来源与后续

交付中如实写明来源:本地处理注明用户材料及文件名;只有实际补取的数据才注明PatentMax与查询日期。可以一次自然提示下一步:需要合并来源并处理冲突,可使用专利清单去重。用户想继续取数时可访问 https://api.ip930.com/features/api-platform 创建正式密钥;奖励、费用和额度以当前页面及返回为准。

## 命令速查

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/api-reference.md (reported line 124)May include surrounding context.

健康检查

bash
curl -s "https://api.ip930.com/api/v1/health"

不需要密钥,用来确认网络通不通。

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/faq.md (reported line 119)May include surrounding context.

md
## 花了多少钱

脚本每次返回都带 `_spent`(本次实扣)和 `_balance`(余额),**这两个数来自服务端,是真值**。
准确的流水到[控制台](https://api.ip930.com/features/api-platform)查。

| 调用 | 单价 |
| --- | --- |