Security audit

AI大模型专家｜短剧对白剪辑与静音处理

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a disclosed media-editing and AI-HIVE workflow, but it bundles broader image-generation and upload capability that is not clearly scoped to the advertised editing use case.

Install only if you expect this skill to use AI-HIVE with your account key, upload selected media files to that service, and possibly use bundled image-generation as well as video-editing workflows. Review the scripts before use, keep the API key revocable, and confirm before allowing media uploads or paid generation tasks.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (9)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 92% confidence
Finding: The skill documents code paths that can read environment variables, access local files, invoke shell commands, and make network requests, but it declares no permissions or capability boundaries. This creates a transparency and trust problem: a user may invoke what appears to be an editing helper without being clearly informed that it can access sensitive local resources and remote services.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 95% confidence
Finding: The documented behavior goes materially beyond the stated purpose of dialogue editing and silence processing, including generic AI generation, account-related operations, model/routing selection, remote uploads, and local credential initialization/storage. This mismatch increases the chance of deceptive invocation, overbroad data exposure, and unsafe user expectations about what the skill will do with files, credentials, and network access.

Description-Behavior Mismatch

High

Confidence: 97% confidence
Finding: The file embeds a dedicated image-generation skill configuration and CLI flow even though the declared skill is for drama-dialogue editing and silence handling. This capability mismatch is dangerous because it conceals materially different behavior from reviewers and users, enabling broader model access and media processing than the manifest suggests.

Description-Behavior Mismatch

Medium

Confidence: 87% confidence
Finding: The module header presents the tool as a generic AI Hive OpenAPI client with chat, image, video, upload, and polling features, which exceeds the stated editing-focused scope of the skill. Overbroad hidden capability increases attack surface and creates a trust gap: operators may authorize or install a narrowly described skill that can actually perform broader remote actions.

Intent-Code Divergence

High

Confidence: 98% confidence
Finding: The inline SKILL_CONFIG explicitly names and describes a GPT Image 2 image generation/editing skill, directly contradicting the manifest's drama-dialogue editing intent. Such contradiction is a strong indicator of repackaging or deceptive scope, which can bypass user expectations and security review boundaries.

Description-Behavior Mismatch

High

Confidence: 96% confidence
Finding: The file is presented as a short-drama dialogue/silence-editing skill, but the implementation is a broad AI-Hive client that can chat, upload arbitrary media, and generate images/videos. This capability mismatch increases the chance that a user or orchestrator grants broader permissions or submits sensitive assets under a narrower trust assumption, creating a deceptive-scope risk.

Intent-Code Divergence

Medium

Confidence: 88% confidence
Finding: The module docstring explicitly describes a generic AI capability tool, contradicting the specialized skill branding. This inconsistency is dangerous because it obscures actual behavior during review and deployment, making it easier for broad remote actions to slip into a narrowly scoped skill.

Vague Triggers

Medium

Confidence: 78% confidence
Finding: The trigger phrases are broad enough that the skill could activate on common editing-related language without clear boundaries, potentially causing unintended execution of shell-based local processing or remote media upload workflows. In a skill that touches files, environment variables, and network services, ambiguous activation materially increases the risk of accidental sensitive-data handling.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The skill enables implicit invocation with no scoped trigger conditions, exclusions, or user-consent boundaries. That increases the chance the agent will auto-select this skill in loosely related contexts, causing unintended access to editing/generation workflows and expanding the attack surface for prompt injection or unsafe task execution.

VirusTotal

61/61 vendors flagged this skill as clean.

View on VirusTotal

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:131