Security audit

Contract Risk Analyzer

Security checks across malware telemetry and agentic risk

Overview

This is a coherent contract-review skill, but users should treat uploaded contracts and API keys as sensitive because the workflow uses external AI services and temporary local files.

Install only if you are allowed to process the contract with the configured AI provider. Use a dedicated low-privilege API key, provide a paid-tier token only when needed, confirm the exact PDF before running analysis, and delete /tmp/contracts outputs after use for confidential documents.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (10)

Vague Triggers

Medium

Confidence: 87% confidence
Finding: The trigger description "当用户触发‘合同审查’时" is overly broad and lacks clear invocation boundaries, which can cause accidental activation during ordinary conversation. In an agent setting, ambiguous triggers can lead to unintended processing of sensitive contract files and downstream actions such as storage and external analysis without sufficiently deliberate user intent.

Missing User Warnings

High

Confidence: 96% confidence
Finding: The README describes analysis via an OpenAI-compatible external API endpoint but does not clearly disclose to users that contract contents may be transmitted to third-party services for processing. Because contracts commonly contain confidential business or personal data, undisclosed external transmission creates significant privacy, compliance, and data-handling risk.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The workflow instructs saving uploaded contracts to /tmp/contracts/<uuid>.pdf without notifying users or describing storage duration, permissions, or cleanup behavior. Local persistence of sensitive legal documents, even in a temporary directory, increases risk of unauthorized access, forensic recovery, or accidental reuse on shared systems.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The skill instructs downloading uploaded contract PDFs to local storage under /tmp/contracts/ and saving generated reports, but it does not require a clear user-facing privacy notice or explicit consent flow for handling sensitive legal documents. Contracts commonly contain confidential commercial terms, personal data, signatures, and dispute details, so silent local storage increases privacy and data-handling risk if the host is shared, logs are accessible, or cleanup is inconsistent.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The workflow sends extracted contract text to an external AI model/API, including potentially the last 8000 characters of the contract, without explicitly warning the user that confidential contract contents may leave the local environment. This is dangerous because contracts often contain trade secrets, pricing, identities, and legal obligations, and external transmission can create confidentiality, compliance, and third-party retention risks.

Vague Triggers

Medium

Confidence: 85% confidence
Finding: The trigger condition is very broad: sending the phrase “合同审查” or uploading a PDF may activate the skill during ordinary conversation or when users share files without intending a contract analysis workflow. In a document-processing skill, broad triggers can cause unintended handling of sensitive legal documents and accidental transmission to OCR/LLM backends.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The skill description encourages users to upload contracts, which commonly contain highly sensitive personal, financial, and commercial data, but it does not clearly warn about privacy, external API transmission, retention, or OCR/LLM processing. This omission can mislead users into exposing confidential content without informed consent, especially because the tool explicitly uses OpenAI-compatible endpoints and OCR processing.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The script sends extracted contract text, which may contain confidential business terms, personal data, or legal secrets, to an external AI provider without explicit user notice or consent. In a contract-review skill, this is materially dangerous because users are likely to process highly sensitive documents and may not expect third-party disclosure.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The provided API key is transmitted to a remote verification service without clear disclosure to the user. Although verification is an intended feature, silently forwarding credentials to a third party can violate user expectations and expose secrets to unnecessary external handling.

Ssd 3

Medium

Confidence: 95% confidence
Finding: Raw contract text is embedded directly into the LLM prompt without delimiting it as untrusted document content or instructing the model to ignore embedded instructions. A maliciously crafted contract could include prompt-injection text that manipulates the analysis, suppresses risks, or causes malformed output, undermining the integrity of the review.

VirusTotal

No VirusTotal findings

View on VirusTotal

Static analysis

No suspicious patterns detected.