Back to skill

Security audit

其他综合收益·逐条挂源

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report behavior and validation concerns beyond its stated financial-disclosure purpose; coverage was partial.

Consider the suspicious verdict before installing. The scan reports an unvalidated destination for a request carrying an API key, external parsing of attachments, a local usage marker, and a checker validation bypass. Skillspector coverage was partial: three of four components were scanned, and static analysis of the fixture shell script was degraded, so the evidence is incomplete.

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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所提供代码仅执行远程目录获取、本地保存及 JSON 解析输出,没有读取财报、提取其他综合收益披露或建立原句追溯登记表的逻辑。该网络请求的资源与声明用途没有可见关联,因此不属于已声明任务的合理支持性实现细节。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所给代码的实际功能是外部 MCP 服务发现与工具枚举,没有读取财报、提取其他综合收益披露或生成可回溯登记表的逻辑。其网络资源访问和令牌认证行为也未在声明中说明。虽然工具发现可能作为更大系统的辅助步骤,但该片段没有展示这些操作与声明的财报整理任务之间的联系,因此存在实质性不匹配。代码仅打印获取到的 routing 信息,并未将其自动赋值给 ROUTING_URL;后续请求的实际目标取决于已有环境变量。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The validation checks reasonably support the declared disclosure-registration workflow. However, mark_first_run() performs an unrelated persistent filesystem write before document processing, including for --demo, --selftest, missing input, or unreadable input. The marker records the skill identity and execution date, and failures are silently suppressed. This usage-tracking behavior is not disclosed in the declared purpose. No network transmission or financial-data recalculation is present.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
90% confidence
Finding

The anonymous catalog GET is not itself sensitive exfiltration, but the surrounding workflow instructs the agent to send CUE_API_KEY to a routing URL obtained from that remote catalog without specifying destination validation. A compromised catalog or malicious routing response could therefore redirect the bearer credential to an attacker-controlled endpoint.

Content

Scanner excerpt · README.md (reported line 84)May include surrounding context.

发现先行,匿名取 live 目录,拿到 routing 后再连接发 tools/list,以现场返回为准,不硬编码工具名与工具数:

bash
curl -sS https://cuecue.cn/api/mcp-catalog -o /tmp/mcp-catalog.json
python3 -c "import json;d=json.load(open('/tmp/mcp-catalog.json'));print(d.get('routing') or d)"
# 拿到 routing 后,带 token 发 tools/list:
curl -sS -X POST "$ROUTING_URL" -H "Authorization: Bearer $CUE_API_KEY" \

The documented data-channel commands issue HTTP requests despite the explicit claim that this skill does not send requests.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

L099 states “本件不发请求” (this skill does not send requests), but L084 fetches a remote catalog and L087–L089 send an authenticated JSON-RPC request to a routing endpoint. Following the supplied workflow therefore performs network activity, including transmitting an API credential, contrary to the stated no-request boundary.

Content

No source excerpt is available for this finding.

The workflow sends financial attachments to an external parsing service without a privacy warning.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

The workflow directs long attachments and scans to an external parser without requiring confirmation that their contents may be disclosed to that provider. Published reports reduce the risk, but user-supplied financial attachments can contain confidential information, and the cost warning does not address confidentiality.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The workflow sends attachments to an external parsing service without warning users about disclosure of document contents.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The skill directs external parsing when evidence resides in attachments, without requiring user authorization to disclose their contents. Although public annual reports are typical inputs, the skill also accepts user-supplied files and screenshots that may contain confidential financial information.

Content

No source excerpt is available for this finding.

Self-calculated entries are exempted from the quotation and amount checks despite the manifest's mandatory verbatim-disclosure policy.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The quotation, tax, and reclassification checks skip an entire finding line whenever it contains 〔自算〕, allowing an untrusted report author to exempt unsupported amounts from validation simply by adding that marker. This directly undermines the skill's requirement that tax-net and reclassification amounts be copied from disclosures rather than calculated, and can produce a zero-findings result for financially misleading entries.

Content

No source excerpt is available for this finding.

The checker silently records daily usage in the user's home directory, unrelated to OCI disclosure analysis.

Low
Category
Excessive Agency
Confidence
99% confidence
Finding

mark_first_run() silently creates a dated usage marker under ~/.cue-usage after argument parsing, including for demo and self-test requests, without user consent. This introduces unrelated persistent state and reveals which skill ran on which dates to anyone able to read that directory; however, the provided code does not transmit the markers or record financial-document contents, so the privacy impact is limited.

Content

No source excerpt is available for this finding.