Security audit

AI领域重点企业资讯抓取与简报生成

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent AI-news collection and briefing tool, with expected RSS/web access and optional third-party AI summarization that users should enable only for shareable content.

Install only if you are comfortable with the skill fetching configured RSS/web sources and, when ARK_API_KEY or an Ark API key is provided, sending article content plus related company/category context to Volcengine Ark/Doubao for summarization. Keep the API key in the environment, avoid processing confidential or regulated material, and review generated summaries for prompt-injection or misclassification effects.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Context-Inappropriate Capability

Medium

Confidence: 94% confidence
Finding: The workflow transmits collected news items to an external AI service using a configured API key and base URL, but there is no visible consent check, data-minimization step, or manifest justification for third-party sharing. Even if the content is mostly public news, the pipeline may include proprietary source selections, annotations, classifications, or internal processing metadata that should not be sent off-platform by default.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The README recommends enabling Doubao/Volcengine summarization via an API key but does not warn that article text, titles, metadata, or derived content may be sent to a third-party LLM provider. In a news aggregation workflow that stores and processes scraped content, this omission can cause unintentional external data disclosure, especially if operators assume all processing remains local.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill description states it collects news from RSS and web pages and can generate summaries via Doubao, but it does not clearly warn users that running the skill will make outbound network requests and may send collected article content or excerpts to a third-party API. This can create privacy, compliance, and data-handling risks because users may provide internal source lists or process sensitive content without realizing it will be transmitted externally.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: This code sends article text, summaries, titles, category labels, and related company metadata to a third-party LLM endpoint without any visible consent gate, redaction step, or policy enforcement in this path. If news items contain proprietary, embargoed, regulated, or personal data, the service can exfiltrate that content to an external processor, creating confidentiality and compliance risk.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: The DeepTranslatorBackend sends arbitrary item text to an external translation provider, which can disclose titles, summaries, and full cleaned content to a third party without any consent, notice, or policy gate in this code path. In a news pipeline this may expose copyrighted, private, embargoed, or otherwise sensitive content, and the service is selected automatically by default when the dependency is installed.

Ssd 1

Medium

Confidence: 96% confidence
Finding: Untrusted article content is concatenated directly into the same prompt that contains task instructions, so malicious text inside a news item can attempt to override the summarization rules, manipulate JSON fields, or induce refusal/malformed output. In this context the model output is written back into application data fields, so prompt injection can corrupt downstream records, misclassify events, or cause repeated processing failures.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.