Security audit

Seedance 视频编辑与延长｜AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed AI-HIVE video editing helper, with review notes around broad activation and API-key storage but no artifact-backed malicious behavior.

Install only if you intend to use AI-HIVE for Seedance video editing or extension. Review prompts and routing before any paid generation, upload only media you are authorized to use, keep the API key out of logs and repositories, and avoid setting AI_HIVE_BASE_URL or --base-url unless you trust the endpoint.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection

Findings (9)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 93% confidence
Finding: The skill includes executable examples and operational guidance that use environment variables, local file paths, network access, and shell commands, but it declares no permissions or capability boundaries. This creates a confused-deputy risk where an agent may read/write files, access secrets, or call external APIs without explicit user awareness or policy gating, especially because the workflow includes uploading media and polling remote services.

Description-Behavior Mismatch

High

Confidence: 98% confidence
Finding: The file is implemented as a generic AI Hive client supporting chat, image generation, video generation, model enumeration, uploads, and task polling, while the declared skill is narrowly scoped to Seedance video editing/extension. This capability mismatch expands what an invoking agent can do far beyond the advertised purpose, increasing the risk of policy bypass, unintended data handling, and misuse of unrelated AI functions under the trust boundary of this skill.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: The skill exposes a user-info endpoint that retrieves account and wallet/balance information even though that data is not necessary to perform Seedance video editing or extension. Unnecessary access to sensitive account metadata violates least privilege and could leak billing or account details to callers that only expected media-processing behavior.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: The chat_text endpoint and corresponding CLI command provide a general text/multimodal chat capability unrelated to Seedance video extension. In skill ecosystems, unrelated broad capabilities are dangerous because they let a supposedly narrow media-editing skill act as a generic model access proxy, potentially processing arbitrary prompts and attached images outside intended review and control paths.

Context-Inappropriate Capability

High

Confidence: 97% confidence
Finding: Standalone image generation is outside the stated Seedance video editing/extension purpose and meaningfully broadens the skill into a more general content-generation tool. That mismatch can be exploited to generate unrelated assets, bypass workflow restrictions, or process prompts that users and reviewers would not expect this skill to handle.

Intent-Code Divergence

Medium

Confidence: 84% confidence
Finding: The module docstring and CLI framing describe a generic AI capability invocation tool, contradicting the metadata that presents this as a Seedance-specific editing skill. This kind of misleading scope declaration is dangerous because operators may grant trust, permissions, or deployment approval based on the narrower description while the code supports materially broader actions.

Vague Triggers

Medium

Confidence: 80% confidence
Finding: The trigger phrases include very broad terms such as '视频修改', '视频延长', and 'Extend Video', which can match many unrelated user requests outside the intended Seedance/AI-HIVE workflow. Overbroad activation can route users into a skill that encourages external uploads, editing operations, and generative workflows when a simpler or safer tool would be more appropriate.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The default prompt embeds a broad invocation phrase that can activate the skill whenever a request is loosely related to video editing, extension, or production-ready workflows. In a skill that can generate runnable AI-HIVE commands and operational workflows, this increases the chance of unintended invocation, causing the model to steer user requests into this skill without clear consent or narrow task matching.

Natural-Language Policy Violations

Medium

Confidence: 96% confidence
Finding: The default prompt forces a Chinese workflow output regardless of the user's language preference. This can override user intent, reduce transparency, and increase the risk that users misunderstand generated commands or operational steps, especially when the skill produces runnable instructions tied to external APIs and production workflows.

Static analysis

No suspicious patterns detected.