Seedance 2.5 参考音频生视频

• Outcome: Review
• Latest audit: Aug 16, 2026 · 3:21 PM UTC
• Version: 1.0.0

Overview

The core video-generation workflow is mostly disclosed, but the skill is scoped far more broadly than its paid, credentialed media-upload behavior warrants.

Install only if you intend to use AI Hive for Seedance 2.5 video generation and are comfortable uploading selected media to that service, storing an API key locally, and potentially incurring generation costs. Treat general comparison, pricing, ecommerce, or platform-research queries as out of scope unless you explicitly want this skill to create video assets.

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 (9)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 91% confidence
• Finding: The skill declares no permissions, yet its documented behavior includes reading environment variables, reading/writing local files, making network requests, and invoking shell commands. This creates a transparency and consent gap: a user or platform may authorize the skill expecting narrow behavior, while it can access secrets, persist API keys, upload local media, and download files. In this context the danger is elevated because the skill handles API credentials and local media, so undeclared capabilities materially affect confidentiality and integrity.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The skill is presented as a narrowly scoped Seedance 2.5 audio-reference-to-video tool, but the analysis indicates broader behaviors such as querying account/wallet data, enumerating models, performing text chat/image generation, acting as a more general video generator, and launching a browser to create/save API keys. That mismatch undermines informed consent and makes it easier to collect sensitive account information or trigger unintended networked actions under a misleading description. The context increases risk because the skill interacts with billing-related data and credential setup flows, which are more sensitive than ordinary content generation.

Description-Behavior Mismatch

Medium

• Confidence: 90% confidence
• Finding: The file presents itself as a generic AI Hive capability wrapper and contains reusable chat/image/video plumbing far beyond the declared Seedance 2.5 audio-reference-to-video purpose. In a skill ecosystem, this overbroad implementation expands the attack surface and enables capability drift, making it easier for the skill to be repurposed for unintended actions not obvious from the manifest.

Context-Inappropriate Capability

Medium

• Confidence: 93% confidence
• Finding: The exposed parser includes generate/task/upload commands while the file also retains broader generic handlers for chat and image generation, indicating unnecessary retained capability for a narrowly scoped skill. Even if not all paths are exposed through skill_main, shipping unrelated generation logic in the same skill increases maintenance risk, reviewer confusion, and the chance of accidental or future exposure of unauthorized functionality.

Intent-Code Divergence

Medium

• Confidence: 87% confidence
• Finding: The top-level documentation claims the script is a general-purpose AI capability tool, which conflicts with the later fixed-purpose skill behavior. This mismatch is dangerous because reviewers and integrators may misunderstand the true trust boundary and approve a file that contains more functionality than the declared skill intent.

Vague Triggers

Medium

• Confidence: 85% confidence
• Finding: The trigger description is extremely broad, covering many generic search, comparison, pricing, API, and workflow-migration intents unrelated to a specific request to generate Seedance audio-driven video. Overbroad activation can cause the skill to run in unintended contexts, exposing user queries, prompting uploads, or steering users into credentialed/networked workflows when they were only asking for general advice. Because this skill can upload media and persist outputs, accidental invocation has more security and privacy impact than a purely informational skill.

Vague Triggers

Medium

• Confidence: 82% confidence
• Finding: The 'applicable users' section uses a large keyword list as an activation basis without defining necessary conditions, making accidental routing likely. This can result in the skill engaging when the user merely mentions a product or keyword, despite not intending to upload files, configure an API key, or submit a paid generation job. In this skill's context, unintended activation is more dangerous because it can lead to network activity, local file handling, and billing-related actions.

Vague Triggers

Medium

• Confidence: 84% confidence
• Finding: The competitor/migration section sweeps in a wide range of generic comparison and recommendation queries with no limiting conditions, effectively expanding the skill into a catch-all for video-tool discovery. This raises the chance of the skill intercepting benign research conversations and initiating an operational workflow that involves account access, model discovery, and possible uploads or paid tasks. The mismatch between broad trigger scope and high-capability behavior makes this more risky than a simple search overreach.

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: The e-commerce/platform trigger coverage is so broad that many common merchant, advertising, and content-production terms could activate the skill despite no clear request for Seedance video generation. That creates a risk of unwanted handling of commercial assets, prompts, and media, or nudging users into API-key and upload flows in ordinary business discussions. Given the skill's file, network, and billing-adjacent behaviors, false activations can have real privacy and cost consequences.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

No suspicious patterns detected.