Back to skill

Security audit

基金资料卡

Security checks for vulnerabilities and agentic risk

Overview

The skill fits its public fund-document summarization purpose, with optional online services that require care around credentials and document privacy.

Use supplied-text mode for private or annotated documents. Before enabling Cue, verify the HTTPS destination receiving your API key and review the parsing provider's privacy and billing terms. Keep keys out of chat, review optional companion skills separately, and prefer reviewed installation versions. Remote service behavior and companion implementations were not verified by this bundle review.

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
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (17)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所给代码仅展示外部服务目录访问及带认证的 MCP 工具枚举,没有读取基金文件、提取八项资料、生成原文引用或标注缺失项的逻辑。代码也未说明这些远程服务如何支持声明的基金资料卡功能,因此其实际用途是连接诊断与工具发现,与声明用途存在实质差异。API 密钥作为认证头发送给 ROUTING_URL 指定的端点;仅凭该片段不能认定存在恶意密钥外传。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所示代码的实际功能是检查 MCP 目录可访问性,并探测远程端点是否接受 API 密钥。代码未读取基金资料、提取声明的八项信息、生成逐字引用或标注来源,也未体现这些网络请求与基金资料处理的必要关联。因此,其主要行为及访问资源与声明用途存在实质性不符。

Content

No source excerpt is available for this finding.

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 · README.en.md (reported line 27)May include surrounding context.

3. Install and run

bash
mkdir -p ~/.workbuddy/skills
cp -r fund-facts-cn ~/.workbuddy/skills/
# ask the agent for the five-section card defined in SKILL.md, save it as my-fund-card.md
python3 ~/.workbuddy/skills/fund-facts-cn/scripts/check_fund_card.py ./my-fund-card.md

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 · README.md (reported line 28)May include surrounding context.

bash
# 1. 放到技能目录
mkdir -p ~/.workbuddy/skills
cp -r fund-facts-cn ~/.workbuddy/skills/

# 2. 让 agent 按 SKILL.md 的基金卡五段出卡,产物存成我的基金卡.md

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The installation command executes an unpinned npm CLI and retrieves skill content from a mutable repository reference. Although this is not itself an MCP server declaration, a compromised upstream release or repository could introduce code or agent instructions with the installing user's privileges.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The optional collection installer uses an unpinned npm executable and mutable repository content. An upstream compromise could therefore replace the expected dependencies with malicious code or agent instructions; the optional nature of installation limits exposure but does not remove the risk.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

Selecting a single skill reduces installation scope but does not pin either the npm installer or the remote skill revision. Changes to either upstream source could execute malicious code or introduce hostile instructions during installation and subsequent use.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

Sending an API key to its intended service is normal authentication, and this request does not upload a fund document. However, the instructions take the destination from a live catalog and send the bearer key without specifying HTTPS, origin, or credential-audience validation; a compromised or misconfigured routing entry could receive the key.

Content

Scanner excerpt · README.en.md (reported line 55)May include surrounding context.

2) take routing.url and transport for the domain from the response; never hard-code the host

3) with the key, send tools/list - proves that endpoint accepted the key, nothing else

curl -s -X POST "$ROUTING_URL" -H "Authorization: Bearer $CUE_API_KEY"
-H "Content-Type: application/json" -H "Accept: application/json, text/event-stream"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | head -c 400

text

The instruction to use third-party attachment parsing lacks a document-sharing privacy warning.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The troubleshooting workflow recommends external attachment parsing without explaining that document contents may leave the local environment or requiring approval. Most intended inputs are public fund documents, reducing sensitivity, but user-provided attachments could contain private annotations or other nonpublic information.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

This installation command runs an unpinned npm CLI and obtains the skill from a mutable repository reference. The scanner's MCP terminology is imprecise, but the underlying supply-chain risk is real: upstream changes could introduce code or agent instructions with the user's privileges.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The collection installation flow leaves both the npm installer and repository revision unpinned. Interactive selection narrows what gets installed but does not prevent a compromised upstream release from supplying malicious code or skill instructions.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx skills'.

Medium
Category
MCP Rug Pull
Confidence
95% confidence
Finding

The single-skill command still executes the latest resolved installer and retrieves mutable remote content. Restricting the skill name does not protect against compromise of the installer or the selected skill's upstream repository.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
86% confidence
Finding

The tools/list request is ordinary authenticated service discovery, not evidence of document exfiltration. Nevertheless, it attaches the bearer key to a catalog-supplied URL without an explicit HTTPS and trusted-origin check, so a compromised or incorrect routing entry could redirect the credential to an unintended recipient.

Content

Scanner excerpt · README.md (reported line 59)May include surrounding context.

2) 域路由现值:从返回里取该域的 routing.url 与 transport,别把域名抄死

3) 带 key 实连后发 tools/list——只证该数据腿端点被接受,不证别的入口

curl -s -X POST "$ROUTING_URL" -H "Authorization: Bearer $CUE_API_KEY"
-H "Content-Type: application/json" -H "Accept: application/json, text/event-stream"
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}' | head -c 400

text

The attachment-parsing workflow lacks a warning about sharing supplied documents with an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The recovery instructions recommend attachment parsing without a clear disclosure that an external provider may receive the supplied document. Public fund materials are normally low sensitivity, but private annotations or nonpublic content in an uploaded attachment could be disclosed without informed approval.

Content

No source excerpt is available for this finding.

The attachment-parsing workflow lacks a warning that user-supplied document content will be sent to an external service.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The operational instructions route attachment content through an external parsing channel when needed, without requiring disclosure or consent before transmission. Although the intended documents are usually public, user-supplied PDFs can contain private material, and treating parsed content as untrusted instructions does not address upload privacy.

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 attachment-parsing workflow lacks an explicit privacy warning about sending user-provided documents to an external service.

Medium
Category
Output Handling
Confidence
86% confidence
Finding

The workflow directs attachment and PDF parsing through an external service without explicitly explaining the document transfer or requiring authorization before uploading user-supplied material. Although the intended documents are public fund disclosures, supplied files can contain private annotations, account details, or nonpublic material, which could consequently be disclosed to a third party.

Content

No source excerpt is available for this finding.