直播回放爆点切片｜AI-HIVE

• Outcome: Pass
• Latest audit: Aug 24, 2026 · 2:39 PM UTC
• Version: 1.0.0

Overview

This skill is a disclosed AI-HIVE-assisted livestream clipping workflow, but it can upload media and start potentially paid generation tasks so users should approve those actions explicitly.

Install only if you are comfortable sending selected media and prompts to AI-HIVE and possibly paying for generation tasks. Keep API keys out of chats and logs, confirm that livestream/source materials are authorized, and require a parameter and cost review before upload or generation.

SkillSpector

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: 89% confidence
• Finding: The skill advertises executable workflows that use environment variables, file operations, network access, and shell tools, but it does not declare permissions or capability boundaries. This weakens reviewability and consent, and can lead to the agent invoking sensitive actions such as reading local files, calling external APIs, or running ffmpeg/python commands without clear upfront authorization.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 94% confidence
• Finding: The documented purpose is livestream highlight clipping, but the skill content exposes broader capabilities including generic text/image/video generation, model querying, API-key initialization, and general ffmpeg processing. This mismatch can mislead reviewers and users about what the skill may actually do, increasing the risk of unexpected external calls, unintended content generation, cost incurrence, or use outside the approved business scope.

Description-Behavior Mismatch

High

• Confidence: 96% confidence
• Finding: The file header and overall implementation reveal a generic AI-Hive client for chat, image generation, video generation, upload, and task polling rather than a narrowly scoped livestream highlight clipping tool. This capability mismatch is dangerous because agents or users may invoke the skill under the assumption it only performs benign clipping, while it can actually generate arbitrary media and access broader account-backed API functionality.

Description-Behavior Mismatch

High

• Confidence: 97% confidence
• Finding: The skill-specific entrypoint forwards a free-form prompt plus optional media into generic video generation modes (t2v/i2v/r2v) instead of implementing the declared clipping workflow. In context, this makes the skill more dangerous because a user selecting a 'livestream highlight clips' skill is silently given arbitrary media synthesis capability that could be used for deceptive or policy-violating content generation.

Context-Inappropriate Capability

Medium

• Confidence: 89% confidence
• Finding: The skill exposes a user-info endpoint that retrieves user/account information and wallet balance, which is unrelated to livestream clipping. While not inherently malicious, unnecessary account inspection expands data exposure and increases the blast radius if the skill is invoked by an agent that only needed media clipping functionality.

Context-Inappropriate Capability

Medium

• Confidence: 90% confidence
• Finding: The chat_text capability enables generic text and multimodal conversation through the same API key, which is outside the declared purpose of generating livestream highlight clips. In a mismatched skill context, this broadens what an invoking agent can do and may permit unintended prompt routing, data disclosure to third-party models, or abuse of the account quota under a misleading skill label.

Context-Inappropriate Capability

Medium

• Confidence: 90% confidence
• Finding: Standalone image generation is not part of the stated livestream replay clipping workflow, yet the file exposes it as a first-class command. This unnecessary expansion of capability is risky because it enables unrelated content creation using the same trusted skill surface, making abuse and policy evasion easier under a benign-seeming package.

Vague Triggers

Medium

• Confidence: 83% confidence
• Finding: Using a generic trigger such as '直播回放' can cause the skill to activate for requests that are not actually asking for highlight clipping or AI-HIVE-backed media generation. Overbroad activation increases the chance of inappropriate tool selection, unnecessary collection of user materials, or accidental initiation of workflows involving external services and local media processing.

Vague Triggers

Medium

• Confidence: 81% confidence
• Finding: The activation scope expands to broad advertising, ecommerce, short drama, and social-media production scenarios without precise constraints tying those scenarios to the skill's stated clipping purpose. In context, that makes the skill more dangerous because it can be invoked as a general content-production tool, pulling in networked generation and media-processing capabilities that exceed the user's likely intent.

Static analysis

No suspicious patterns detected.