ClawHub
Back to skill

Security audit

NewsToday

Security checks across malware telemetry and agentic risk

Overview

NewsToday is a coherent news briefing skill with disclosed web fetching, local preference memory, and optional scheduled push behavior.

Install this if you want a personalized news briefing skill. Before enabling push, confirm you are comfortable with recurring cron-based notifications and with storing your news preferences in MEMORY.md; use the provided off command and delete the newstoday MEMORY.md profile block to remove that state.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The file header asserts there is no network activity and that all URLs are not influenced by external input, but the generated prompt explicitly instructs a downstream agent to perform WebFetch requests to multiple remote URLs. This mismatch can cause reviewers or calling systems to underestimate the skill’s external reach, weakening trust boundaries and allowing unintended outbound requests to occur under a misleading safety description.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The listed trigger phrases include very generic requests such as 'what happened today', 'what's trending', and 'today's news', which can overlap with ordinary user intent and cause accidental invocation of this skill instead of a more appropriate handler. In a voice- or chat-routed agent ecosystem, overly broad triggers increase the risk of misrouting user requests, unexpected data fetching, and confusing or unauthorized notification behavior.

Vague Triggers

Medium
Confidence
93% confidence
Finding
The invocation examples include very broad phrases like 'what happened today' and 'what's trending' that can match many general queries not specifically asking for this skill's aggregated-news workflow. This can cause unintended routing, overshadow more appropriate skills, and trigger unnecessary fetching or push-related logic when the user's intent is ambiguous.

Vague Triggers

Medium
Confidence
91% confidence
Finding
The detailed Chinese invocation guidance contains highly ambiguous triggers such as '有什么大事' and '今天发生了什么', which are common conversational phrases that may refer to many domains beyond a news brief. In an agent environment, such broad matching increases accidental invocation risk and can override user intent or other better-scoped skills.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The manifest uses very broad, generic activation phrases such as 'daily news brief', 'news digest', 'breaking news alert', and multiple multilingual equivalents that closely match common user requests. In agent ecosystems, overly broad metadata can cause the skill to be invoked for many generic news-related prompts, increasing the chance of unintended routing, user confusion, and shadowing of more appropriate skills.

Session Persistence

Medium
Category
Rogue Agent
Content
push: { enabled: false }
};

const topList = Object.entries(topics).filter(([, w]) => w >= 0.7).map(([t]) => t).join('、');
const topTopicsCsv = Object.entries(topics).filter(([, w]) => w >= 0.7).map(([t]) => t).join(',');

console.log(`
Confidence
87% confidence
Finding
pList

Session Persistence

Medium
Category
Rogue Agent
Content
👤 用户：${userId}
🌐 语言：${language === 'zh' ? '中文' : 'English'}
📌 重点话题：${topList || '默认'}
📡 推送渠道：${channel}
`);
Confidence
83% confidence
Finding
pList

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.