Security audit

合同卫士 / Contract Guardian

Security checks across malware telemetry and agentic risk

Overview

This skill coherently provides local contract review, risk checks, comparison, and expiry tracking, with the main privacy consideration being local processing and storage of contract data.

Install only if you are comfortable letting a local agent process contract contents. Use explicit file paths, avoid pointing it at unrelated private files, and only archive contracts whose metadata you want retained under CG_DATA_DIR.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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

Findings (6)

Vague Triggers

Medium

Confidence: 89% confidence
Finding: “或类似意图”这类开放式触发条件会使合同审查流程在语义相近但并非明确授权的场景下被触发。由于该流程会读取用户指定文件并分析内容，误触发可能导致不必要的本地文件访问或对敏感合同进行处理。

Vague Triggers

Medium

Confidence: 88% confidence
Finding: 风险专项分析同样使用宽泛的“或类似意图”，缺乏明确边界，容易把普通咨询、示例讨论或非目标文本误判为需要执行分析。结合文件读取/文本处理能力，这会扩大敏感合同内容被处理或展示摘要的机会。

Vague Triggers

Medium

Confidence: 87% confidence
Finding: 到期提醒流程的开放式触发可能在用户只是泛泛询问合同状态时就触发本地存档查询。由于该操作会读取已归档合同信息，误触发会带来本地敏感元数据暴露风险。

Vague Triggers

Medium

Confidence: 90% confidence
Finding: 合同存档流程会触发信息提取并写入本地存储，而宽泛的触发描述可能使系统在用户未明确授权持久化时执行归档。相比只读分析，这类误触发还涉及数据落盘，风险更高。

Vague Triggers

Low

Confidence: 82% confidence
Finding: 合同对比流程的触发条件较宽，但其危险性相对较低，因为还包含付费校验和需要两份明确文件路径。尽管如此，模糊触发仍可能导致非预期地读取两份本地文件并生成差异报告。

Missing User Warnings

Low

Confidence: 81% confidence
Finding: Describing paid-feature activation solely through the environment variable `CG_SUBSCRIPTION_TIER=paid` suggests that client-side configuration alone enables entitlement changes. If the skill actually trusts this variable for access control, any user could self-upgrade features without authorization, undermining licensing and potentially exposing premium-only processing paths without server-side validation.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.