Back to skill

Security audit

Omni Reader 多模态文件解析

Security checks for vulnerabilities and agentic risk

Overview

Scanners report a mismatch between the skill’s advertised document parsing and its reported Markdown export behavior, alongside credential-handling, remote-processing, and mutable-install concerns; the verdict is suspicious with medium confidence.

Consider avoiding installation until the advertised capabilities, remote document processing, API-key handling, and package versions are clarified. Do not paste a reusable API key into chat; verify what information is sent to the remote service and whether the commands’ package sources and versions are trustworthy. The scan is incomplete: Skillspector inspected the sole listed component but marked analysis partial because several referenced paths or resources were missing, so the results do not establish that the skill is safe.

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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
Findings (20)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied commands have a materially different primary purpose: exporting Markdown to document formats, which reverses the declared conversion direction. They also explicitly persist generated files. The code does not demonstrate the advertised ingestion and extraction capabilities.

Content

No source excerpt is available for this finding.

Privacy assurances obscure the remote-processing disclosure.

High
Category
Output Handling
Confidence
99% confidence
Finding

Repeated claims that source files do not reach the server, including the FAQ's claim about local mode, conflict with the explicit architecture stating that actual parsing occurs at Cue's remote service. Users could authorize processing confidential documents under a false local-only assumption; no_store concerns retention and does not establish that content is never transmitted or processed remotely.

Content

No source excerpt is available for this finding.

The onboarding instructions solicit an API key in the conversation while presenting the process as user-controlled credential protection.

High
Category
Data Exfiltration
Confidence
99% confidence
Finding

The instructions present key creation as user-controlled protection while requiring the resulting secret to be disclosed in chat. User-only creation does not protect a key after disclosure; this is the same underlying credential-exposure issue identified in the onboarding finding.

Content

No source excerpt is available for this finding.

MCP Config Access

High
Category
Agent Snooping
Confidence
95% confidence
Finding

Skill accesses MCP server configuration files (mcp.json). MCP configs contain server URLs, authentication tokens, and tool definitions — reading them allows the skill to discover and potentially abuse other tool integrations.

Content

Scanner excerpt · SKILL.md (reported line 159)May include surrounding context.

npx -y @cueai/omni-reader-mcp@1.8.6 setup

text

Agent 配置(`~/.claude/mcp.json`):

```json
{

External Script Fetching

High
Category
Supply Chain
Confidence
90% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · SKILL.md (reported line 391)May include surrounding context.

bash
# 健康检查:验证 Key 和连接状态,不输出密钥原文
echo "=== 1/4 API Key ===" && [ -f "$HOME/.cue/config.json" ] && echo "已配置" || echo "未配置!"
echo "=== 2/4 MCP 连接 ===" && curl -sS --max-time 10 -X POST "https://mcp.cuecue.cn/api/omni-reader/mcp/" -H "Content-Type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | python3 -c "import sys,json;r=json.load(sys.stdin);print('已连接 ('+str(len(r.get('result',{}).get('tools',[])))+' 工具)') if 'result' in r else print('连接失败')"
echo "=== 3/4 Node.js ===" && node -v 2>/dev/null && npx -v 2>/dev/null && echo "就绪" || echo "未安装!"
echo "=== 4/4 剩余积分 ===" && python3 -c "
import json, urllib.request, os

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 77)May include surrounding context.

md
> 两者**互补不冲突**:专项 Skill 底层正是调用 Cue Omni Reader 远程端点完成解析(并自动绕过平台反爬),再由大模型加工交付。可直接对我说「安装 xiaohongshu-video-studio」。

## 核心优势                                                                                                                                 
  1. 效果准(官方实测):模糊复杂表格的扫描件解析准确率 高达97.39%,已和主流OvisOCR2(准确率94.77%)、合合 textin(92.16%)、MinerU2.5(91.50%)等6大主流ocr工具做过效果对比,准确率处于行业第一梯队;                                                                                                                                                                     
  2. 隐私友好:默认 no_store 不落盘,源文件与解析结果不上服务端;
  3. 文档大小不限制,支持单文件256MB,300+页pdf文档也是秒级解析速度;支持多文件批量解析,上百份文档解析批量处理。

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · SKILL.md (reported line 78)May include surrounding context.

md
> 两者**互补不冲突**:专项 Skill 底层正是调用 Cue Omni Reader 远程端点完成解析(并自动绕过平台反爬),再由大模型加工交付。可直接对我说「安装 xiaohongshu-video-studio」。

## 核心优势                                                                                                                                 
  1. 效果准(官方实测):模糊复杂表格的扫描件解析准确率 高达97.39%,已和主流OvisOCR2(准确率94.77%)、合合 textin(92.16%)、MinerU2.5(91.50%)等6大主流ocr工具做过效果对比,准确率处于行业第一梯队;                                                                                                                                                                     
  2. 隐私友好:默认 no_store 不落盘,源文件与解析结果不上服务端;
  3. 文档大小不限制,支持单文件256MB,300+页pdf文档也是秒级解析速度;支持多文件批量解析,上百份文档解析批量处理。
  4. 文件类型不限制,支持PDF/Office/图片/扫描件/音视频/网页/压缩包等所有常见文件类型;

API-key onboarding requests disclosure in chat without a credential-handling warning.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

Onboarding explicitly asks the user to paste a reusable API key into the conversation. This unnecessarily exposes the credential to model context and any retained conversation records, where later disclosure could enable unauthorized service use or credit consumption.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @cueai/omni-reader-mcp'.

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The update command uses @latest, which is a mutable reference rather than a pinned version. A compromised future release could execute with the user's privileges and modify agent configuration; requiring upgrade consent reduces automatic exposure but does not verify the downloaded code.

Content

No source excerpt is available for this finding.

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

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Mandatory verbatim Chinese prompts impose an output language without user choice.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The skill requires the balance prompt and options to be used verbatim, explicitly prohibiting changes to their Chinese text. This forces Chinese output even for users communicating in another language, without a language choice or a justified locale constraint.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 366)May include surrounding context.

md
| 目标格式 | 依赖 | macOS | Ubuntu |
|----------|------|-------|--------|
| Word (.docx) | pandoc | `brew install pandoc` | `sudo apt install pandoc` |
| PDF (.pdf) | pandoc + LaTeX | `brew install --cask basictex` | `sudo apt install texlive-xetex` |

---

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 367)May include surrounding context.

md
| 目标格式 | 依赖 | macOS | Ubuntu |
|----------|------|-------|--------|
| Word (.docx) | pandoc | `brew install pandoc` | `sudo apt install pandoc` |
| PDF (.pdf) | pandoc + LaTeX | `brew install --cask basictex` | `sudo apt install texlive-xetex` |

---

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

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @cueai/omni-reader-mcp'.

Medium
Category
MCP Rug Pull
Confidence
99% confidence
Finding

The recovery command runs 'npx -y @cueai/omni-reader-mcp setup' without a version, allowing resolution to a mutable published release. A compromised release would run locally with access available to the agent, and setup is explicitly capable of changing its configuration.

Content

No source excerpt is available for this finding.

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 · SKILL.md (reported line 391)May include surrounding context.

bash
# 健康检查:验证 Key 和连接状态,不输出密钥原文
echo "=== 1/4 API Key ===" && [ -f "$HOME/.cue/config.json" ] && echo "已配置" || echo "未配置!"
echo "=== 2/4 MCP 连接 ===" && curl -sS --max-time 10 -X POST "https://mcp.cuecue.cn/api/omni-reader/mcp/" -H "Content-Type: application/json" -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}' | python3 -c "import sys,json;r=json.load(sys.stdin);print('已连接 ('+str(len(r.get('result',{}).get('tools',[])))+' 工具)') if 'result' in r else print('连接失败')"
echo "=== 3/4 Node.js ===" && node -v 2>/dev/null && npx -v 2>/dev/null && echo "就绪" || echo "未安装!"
echo "=== 4/4 剩余积分 ===" && python3 -c "
import json, urllib.request, os

The privacy documentation says the skill does not read or transmit API keys, but its mandatory health check does both.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The health check reads the configured API key and sends it as a Bearer token to Cue's billing endpoint, contradicting the categorical claim that the skill does not read or transmit keys. This is normal first-party authentication rather than demonstrated credential theft, but the inaccurate assurance prevents informed authorization of the credential's use.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @cueai/omni-reader-mcp'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The FAQ recommends an @latest setup command, reintroducing mutable dependency execution despite pinned installation examples elsewhere. Approval to update does not establish that the newly resolved release is trustworthy.

Content

No source excerpt is available for this finding.

MCP server referenced without pinned version: 'npx -y @cueai/omni-reader-mcp'.

Medium
Category
MCP Rug Pull
Confidence
98% confidence
Finding

The FAQ's unversioned uninstall command can resolve and execute a newly published package before performing removal. Uninstallation therefore remains a supply-chain execution surface rather than an inherently safe cleanup action.

Content

No source excerpt is available for this finding.

Generic parsing requests activate the skill without a clear file-processing scope.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

The activation instructions include generic requests such as “解析一下” (“analyze/parse this”) without requiring an identified document, file, or URL. This can overlap with requests to explain code or analyze an idea rather than invoke a remote document parser.

Content

No source excerpt is available for this finding.