Security audit

Voyage AI CLI

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent Voyage AI and MongoDB Atlas CLI guide, with expected but sensitive API, database, and local configuration behaviors users should handle deliberately.

Before installing, verify the voyageai-cli npm package and publisher. Use least-privilege Voyage AI and MongoDB credentials, avoid putting real API keys directly in shell history, and test store, ingest, index create, and index delete against non-production databases first. Do not embed, rerank, or ingest sensitive documents unless you intend for that content and metadata to be processed by the configured external services and, for storage workflows, persisted in MongoDB Atlas.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (3)

Description-Behavior Mismatch

Low

Confidence: 93% confidence
Finding: The skill exposes persistent local credential/config storage via `vai config set api-key` and related commands, but the manifest/description focuses on Atlas/vector search features without clearly disclosing that secrets may be written to local disk. This can lead users or orchestrators to invoke the skill under the assumption it is stateless, increasing the chance of credential persistence on shared systems or endpoints with weak local protections.

Vague Triggers

Medium

Confidence: 84% confidence
Finding: The trigger list contains broad natural-language phrases such as `similarity search`, `store embeddings`, and `explain embeddings` that may match ordinary user requests and cause unintended invocation. In this skill, accidental invocation is more concerning because commands can call external APIs and write data to MongoDB Atlas, turning an overbroad routing issue into unwanted data egress or persistence.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill description and command guidance do not clearly warn that text provided to commands like `embed`, `rerank`, `store`, `search`, and `ingest` may be sent to external Voyage AI services and may also be persisted in MongoDB Atlas. Without that disclosure, users may provide sensitive content under the mistaken assumption processing is local or ephemeral, creating confidentiality and compliance risks.

VirusTotal

59/59 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.