Back to skill

Security audit

预期信用损失·三阶段与两条计提路径逐条挂源

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marked this skill suspicious because scanners reported capabilities and data handling that do not clearly match its financial-report checking purpose, alongside weaknesses in its validation checks; analysis was partial.

Consider avoiding installation until the credential handling and external document processing are clarified, and the financial validation gaps are addressed. The scan was partial: three of four components were fully inspected, one was partially inspected, and a referenced path was missing; therefore 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
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所给代码的实际功能是 MCP 服务目录获取和带凭据的工具枚举,未体现报告读取、原句溯源、登记表生成或计提比例与滚动轧差机检。虽然工具发现可能作为其他任务的基础设施,但此处没有展示其与声明的财务附注处理之间的关联,因此不能将这些外部资源访问和凭据使用视为已由声明目的涵盖的辅助实现。代码本身不足以证明凭据被恶意窃取,但足以认定存在未声明能力及资源访问不一致。

Content

No source excerpt is available for this finding.

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 136)May include surrounding context.

md
| `scripts/fixtures/run_fixtures.sh` | 离线样例 runner:自测→好样→坏样三档,坏样注入必先断言落地;期望退码由自测末行的「不可判=N」现算,不写死 |

External Transmission

Medium
Category
Data Exfiltration
Confidence
82% confidence
Finding

Fetching a public catalog is not itself sensitive, but the documented workflow subsequently sends CUE_API_KEY to a configurable MCP_ENDPOINT without enforcing HTTPS or an approved destination. If routing data or endpoint configuration is manipulated, the bearer credential can be disclosed to an unintended service.

Content

Scanner excerpt · README.md (reported line 111)May include surrounding context.

bash
  # ① 匿名取域清单与 routing
  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)"

  # ② 带凭据握手取工具表

The workflow directs screenshots and scanned documents to an external reader without warning about third-party disclosure.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The workflow directs an agent to submit screenshots and scans to an external reader without a confidentiality check or explicit disclosure approval. Published financial reports reduce the usual sensitivity, but user-supplied scans can contain annotations or nonpublic information that would leave the user's environment.

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 screenshots and scanned documents to cue-omni-reader without warning that their contents will be shared with an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The skill instructs the agent to route screenshots and scans to cue-omni-reader without explaining the external disclosure or requiring approval. Although the intended reports are generally public, user-provided files may include confidential annotations or nonpublic material.

Content

No source excerpt is available for this finding.

The stage checker does not enforce its documented rejection of a section containing no stage entries.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

check_stage documents rejection of a stage section containing no stage entries, but it never counts entries or emits that finding. A document can retain the required section heading while omitting every stage entry and still pass this gate, making an incomplete financial register appear validated.

Content

No source excerpt is available for this finding.

The documented quotation-to-value consistency check is never performed.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The quotation checker delegates quotation-to-value consistency to ECL-ROLL, but ECL-ROLL only compares extracted registration values against one another and never compares them with quoted values. An altered, internally consistent set of financial figures can therefore pass validation while contradicting its attached quotations, undermining the skill's central traceability guarantee.

Content

No source excerpt is available for this finding.