人生岔路模拟栈

• Outcome: Review
• Latest audit: Jul 13, 2026 · 4:24 AM UTC
• Version: 1.0.3

Overview

This is a coherent Chinese-language life-choice reflection tool that asks for personal context and can generate local reports, but I found no hidden external actions, credential use, or destructive behavior.

Install this if you want a Chinese-language life-choice retrospection and report generator. Treat health, finance, family, employer, school, location, and relationship details as sensitive: share only what is needed, keep identifying details high-level, and review any generated Markdown or HTML before saving or sharing.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (9)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 93% confidence
• Finding: The skill advertises no explicit permissions, yet its instructions reference file reads/writes and shell-executed validation/rendering scripts. In an agent platform, this creates a capability-transparency gap: users and reviewers may invoke a skill believing it is conversation-only, while it is actually designed to operate on repository files and run local tooling.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 90% confidence
• Finding: The declared purpose is end-user life-decision simulation, but the skill also instructs the agent to package archives, validate repository structure, and run subprocess-based tooling across artifacts. That mismatch is dangerous because it expands operational behavior beyond user expectations, increasing the chance of unintended file-system access, shell execution, and use in contexts unrelated to the requested task.

Vague Triggers

Medium

• Confidence: 79% confidence
• Finding: The default prompt embeds a direct activation phrase and immediately drives the model into generating a full HTML report from minimal user input. Because invocation is broad and paired with implicit invocation, the skill can be triggered in contexts the user may not clearly intend, causing unexpected role/format steering and unreviewed disclosure or transformation of user-provided content.

Natural-Language Policy Violations

Medium

• Confidence: 90% confidence
• Finding: The protocol is predominantly written in Chinese and includes mandated Chinese response templates, which can cause the agent to reply in Chinese regardless of the user's language preference. This creates a language-consent and usability issue: users may receive outputs they cannot understand, and important caveats or uncertainty markers may be missed.

Natural-Language Policy Violations

Medium

• Confidence: 96% confidence
• Finding: The file is entirely in Chinese and provides no language-selection or opt-in mechanism, which can cause users who do not read Chinese to misunderstand instructions, outputs, and limitations. In this skill context, that increases the risk of user confusion around report interpretation and decision-making boundaries, especially because the skill produces structured advice-like artifacts intended for sharing and review.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The workflow explicitly elicits sensitive personal data including emotional state, career situation, cash flow, family relationships, health/energy, and external life events, but the file does not require a privacy notice, minimization guidance, or handling restrictions before collection. In a conversational skill, this increases the risk of oversharing, downstream retention, and unintended exposure of highly sensitive user information.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The skill discusses generating HTML output containing user life-history content and only mentions virtualization/de-identification for public-display scenarios, leaving the default reporting flow without a clear privacy safeguard. If rendered, stored, or shared, HTML reports can preserve sensitive personal narratives and identifying context, increasing disclosure risk beyond the chat itself.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: The file is entirely in Chinese and does not offer a language choice or document that the skill is intentionally restricted to Chinese-speaking users. This can cause users, reviewers, or downstream agents operating in other languages to misunderstand instructions, scoring criteria, and safety-relevant constraints, leading to incorrect outputs or unsafe handling of user-facing content.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The intake template explicitly solicits sensitive personal data, including finances, family responsibilities, health status, relationships, and support networks, but provides no privacy notice, minimization guidance, or data-handling boundaries. In an agent skill context, this creates unnecessary exposure of highly sensitive user information that could be retained, logged, or reused beyond the user's expectations.

VirusTotal

53/53 vendors flagged this skill as clean.

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical

• Code: suspicious.dynamic_code_execution
• Location: scripts/validate_html_first_skill.py:453