Pre-install Check: Security Risk Quick Scan for Chinese Agent Tools

The Problem: Security Blind Spots in Chinese Agent Tool Ecosystems

Before installing a Chinese Agent tool (like Skill, MCP, or plugin), how can you assess if it's safe? Many tools from platforms like SkillHub or GitHub lack unified risk visibility. Users often face pain points:

• Excessive permissions: Tools may demand unnecessary system access or data permissions.
• Privacy leakage risks: Code or declarations might secretly upload data to unknown domains.
• Uncertain author reputation: Tools from individuals or small teams are hard to vet.
• Lack of independent assessment: Platform reviews often focus on compatibility rather than security, leaving users without objective references.

Traditional methods like reading code line-by-line are unfriendly to non-developers, and tools update frequently, making manual audits impractical.

Core Capability: How It Works

"Pre-install Check" acts as an independent tool risk query layer to address these issues. It doesn't mirror source platforms but autonomously analyzes public code and declarations to help you achieve "informed consent." Here's how it operates:

Query Entry Points

You can invoke it via two methods:
- Conversational call (recommended): In tool-supported Agents (e.g., WorkBuddy), use /装前查 with natural language queries, such as:
- /装前查 API Gateway
- /装前查 作者 ctz168
- /装前查 高风险 Top20
- CLI call: Using a Python 3 script (standard library only), outputs plain text results.

Risk Assessment Framework

The skill returns a safety rating for quick decision-making:
- High Risk (red): Hits high-risk evidence; confirm permissions and data flow before installation.
- Cautious (orange): Mid-to-high risk signals; review evidence first.
- Observation (yellow): Mild signals; be informed before use.
- No Risks Found (green): No obvious risk signals detected under current rules.
- To Be Evaluated (unknown): Insufficient evidence or evaluation pending.

Each rating is evidence-based, like code proof (e.g., dangerous API calls), documented self-declarations (e.g., permission requests), or high-risk combinations. Results list key signals, for example:
- Tool declares network access and user data uploads.
- Code contains unencrypted remote connections.

Data Sources and Output

When querying, it handles common links:
- SkillHub links: Parses tool name and searches.
- GitHub links: Queries via repository name.
Output includes a summary, privacy direction (e.g., external domains), author-side stats (e.g., number of evaluated tools), and a details page link (e.g., https://www.zhuangqiancha.com/tool/skillhub%3Axxx) for deeper inspection.

Applicable Boundaries and Caveats

"Pre-install Check" is an auxiliary tool, not a security guarantee. Key notes:

• Independent assessment: Based on public info, it doesn't perform dynamic tests; conclusions may be lagging or incomplete.
• Evidence-driven: Ratings rely on existing rules; No Risks Found (green) doesn't mean absolute safety—still recommend authorizing as needed.
• No fabrication: Strictly follows evidence; uninformed tools are reported truthfully.
• Advisory stance: For High Risk or Cautious ratings, suggest delaying or isolated testing; for To Be Evaluated, prefer tools with complete assessments.

In summary, it provides a quick risk lens, but final decisions should combine with your own needs and review.

Pre-install Check evaluates independently based on public code and declarations; conclusions do not constitute guarantees.