Seedance 视频编辑

• Outcome: Pass
• Latest audit: Aug 16, 2026 · 3:36 PM UTC
• Version: 1.0.0

Overview

The skill is a disclosed AI Hive video-editing wrapper that uploads user-specified media, stores an API key locally, submits tasks, polls status, and downloads outputs, with some broad discovery wording users should notice.

Install only if you are comfortable giving AI Hive an API key and uploading selected media files to its service. Use explicit file paths, avoid passing unrelated private files to upload or generate, and treat the broad search/competitor wording as marketing/discovery scope rather than proof of official integrations.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• 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

Findings (8)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 90% confidence
• Finding: The skill documents capabilities that access environment variables, read/write local files, use the network, and invoke shell commands, yet it declares no permissions. This weakens user and platform visibility into what the skill can do and can enable unexpected exfiltration of API keys, local media, or other sensitive data once the skill is invoked.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 94% confidence
• Finding: The declared purpose is narrowly framed as Seedance video editing, but the skill behavior appears to extend into unrelated API functions such as wallet/user queries, generic chat, image generation, broader video generation, and browser-based API-key setup with local config management. This mismatch increases the chance that users or the platform authorize the skill under false assumptions, enabling unnecessary access to account data and broader remote actions than expected.

Description-Behavior Mismatch

Medium

• Confidence: 94% confidence
• Finding: The file is presented as a fixed Seedance video-editing skill, but it embeds a broader AI Hive client that can perform chat, image generation, model enumeration, account inspection, and generic video generation. This creates a capability mismatch: a caller or integrator expecting a narrow video-edit tool may unknowingly expose broader account-scoped actions and data surfaces than intended.

Context-Inappropriate Capability

Medium

• Confidence: 89% confidence
• Finding: Exposing `user-info` and `models` allows the skill to inspect account metadata, wallet/balance information, and available model inventory outside the stated video-editing workflow. In a delegated agent environment, that expands what the skill can reveal or act upon beyond user expectations and violates least privilege.

Context-Inappropriate Capability

Low

• Confidence: 82% confidence
• Finding: A standalone upload command permits arbitrary file transfer to the remote service independently of any Seedance edit task. While not inherently malicious, it broadens the skill from 'edit a supplied video' into a generic exfiltration-capable uploader, which is risky in agent contexts handling sensitive local files.

Vague Triggers

Medium

• Confidence: 85% confidence
• Finding: The skill description contains extremely broad trigger phrases spanning many generic video, advertising, e-commerce, and competitor-tool queries. Overbroad activation can cause the skill to be selected in contexts where users did not intend to grant file, network, or account-related operations, increasing the risk of surprise execution and data exposure.

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: The applicable-user section uses generic search terms without clear boundaries, making invocation scope ambiguous. In a skill that can touch local files, network services, and credentials, ambiguous routing increases the chance of accidental activation and unintended access to user assets or account operations.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The related-tools section claims suitability for broad comparison, migration, and alternative-product searches across many products without clear trigger limits. This expands the skill’s operational footprint far beyond its core function and can route unrelated user queries into a skill that performs uploads, remote API actions, task polling, and local file writes.

VirusTotal

65/65 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.