AI大模型专家｜DeepSeek AI搜索收录与引用

• Outcome: Review
• Latest audit: Aug 24, 2026 · 9:09 AM UTC
• Version: 1.0.0

Overview

This skill is a disclosed AI-HIVE workflow helper that can use a local API key, upload user-selected media, and run image/video generation, but I did not find hidden, destructive, or deceptive behavior.

Install only if you intend to use AI-HIVE for planning plus media-generation workflows. Treat prompts and uploaded images, videos, or audio as data sent to AI-HIVE, and expect generation tasks to potentially affect account usage or cost. Use authorized media, keep the API key local and revocable, and prefer reviewing commands before running uploads or generation.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• 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
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (7)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 88% confidence
• Finding: The skill advertises executable scripts that use environment variables, shell commands, local file reads/writes, and network access, but the manifest does not declare those capabilities or permissions. This creates a trust and review gap: users may run code that can access local data or send data off-host without having been clearly warned in the skill metadata.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 94% confidence
• Finding: The documented purpose focuses on search inclusion/citation planning, but the referenced behavior extends to account queries, wallet/balance access, browser-based API key setup, generic chat calls, file uploads, and media generation/downloads. This mismatch is dangerous because users may grant trust based on a narrow planning description while the skill can perform broader actions involving credentials, billing-sensitive operations, and data exfiltration to third-party services.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The implementation materially diverges from the declared skill purpose: instead of a DeepSeek search/citation advisory tool, it is a general AI-Hive client exposing chat, image/video generation, uploads, account bootstrap, and task polling. This creates a capability mismatch that can mislead users into granting credentials and running networked operations they did not intend, increasing the risk of unauthorized external service use and data transfer.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The embedded skill configuration hardwires a GPT Image 2 generation/editing workflow, which is unrelated to the advertised DeepSeek search/citation expertise. Because the concrete behavior is hidden behind misleading metadata, users may invoke the skill expecting analysis but instead trigger media generation and outbound API activity, a classic deceptive-capability issue.

Context-Inappropriate Capability

Medium

• Confidence: 97% confidence
• Finding: The init flow opens a browser and walks the user through logging into AI-Hive and creating/copying an API key, despite the stated skill purpose not requiring such account linking. In context, this is dangerous because a misleading skill can socially engineer users into provisioning third-party credentials and persisting them locally for an unrelated service.

Description-Behavior Mismatch

Medium

• Confidence: 94% confidence
• Finding: The file implements a broad AI Hive client with account lookup, model listing, media upload, and generation capabilities, which exceeds the declared skill purpose of DeepSeek search/citation planning. This creates a capability mismatch that can mislead users and grant the skill powers unrelated to its manifest, increasing the attack surface and the chance of unintended data transfer or costly API actions.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The embedded SKILL_CONFIG and skill_main logic repurpose the file into a short-drama video generator, which is materially inconsistent with the published DeepSeek search/citation expert description. Such hidden or mismatched functionality is dangerous because users may authorize or invoke the skill expecting planning/search assistance while it actually performs media uploads, paid video generation, and task polling against a third-party service.

VirusTotal

65/65 vendors flagged this skill as clean.

Static analysis

Detected: suspicious.exposed_secret_literal

File appears to expose a hardcoded API secret or token.

Critical

• Code: suspicious.exposed_secret_literal
• Location: SKILL.md:133