AI大模型专家｜短剧分集梗概

• Outcome: Review
• Latest audit: Aug 24, 2026 · 9:19 AM UTC
• Version: 1.0.0

Overview

This is a disclosed AI-HIVE short-drama planning and media-generation skill, but it uses remote APIs, an API key, and user-selected file uploads.

Install only if you intend to use AI-HIVE for short-drama planning plus image/video generation. Treat it as a remote API client: review files before uploading them, protect the API key stored in ~/.ai-hive/config.json or AI_HIVE_API_KEY, and expect generated files to be downloaded locally when --no-download is not used.

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: 83% confidence
• Finding: The skill advertises executable scripts and capabilities including shell use, environment-variable access, local file read/write, and network communication, but does not declare permissions. This weakens user trust boundaries and informed consent: users may invoke setup or generation commands without realizing the skill can access local configuration, write files, or transmit data to remote services.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 95% confidence
• Finding: The documented purpose is a narrow pre-production/planning skill for episode synopses, but the referenced scripts appear to implement a much broader AI-HIVE client with API-key initialization, model discovery, chat calls, media upload, image/video generation, task polling, and result download. This mismatch is dangerous because users may trust and run it for harmless planning while it actually enables broad remote actions and data transfer unrelated to the stated scope.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The file presents itself as an AI Hive OpenAPI media toolkit with chat, image, video, upload, and polling capabilities, while the published skill metadata describes a drama episode synopsis assistant. This functionality mismatch is dangerous because it expands the operational scope far beyond what a user or reviewer would expect, enabling network access, file upload, and model operations under a misleading label.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The embedded SKILL_CONFIG hardcodes an image-generation/editing skill targeting GPT Image 2, which directly conflicts with the advertised episode-synopsis use case. This is risky because hidden or mislabeled capability changes can cause users to provide secrets or files to a tool they did not intend to trust for media generation and uploads.

Description-Behavior Mismatch

High

• Confidence: 98% confidence
• Finding: The file claims to be a short-drama episode-synopsis skill, but the implementation is a broad AI-Hive client that can upload media, query account data, list models, and generate images/videos. This is dangerous because users and security reviewers may grant permissions or trust based on the advertised synopsis-only purpose while the code enables materially broader actions and external data transfer.

Description-Behavior Mismatch

High

• Confidence: 99% confidence
• Finding: The embedded SKILL_CONFIG hardcodes a 'short-drama-video' video-generation skill, which directly contradicts the metadata describing an episode-synopsis tool. This mismatch increases the risk of deceptive capability exposure, where a user expecting text planning functionality instead runs a tool that uploads files and invokes external video-generation services.

Context-Inappropriate Capability

Medium

• Confidence: 90% confidence
• Finding: User-info retrieval, including wallet balance, is unrelated to generating episode synopses and expands access to sensitive account metadata without clear need. In the context of a mismatched skill, this broadening is more suspicious because it collects information beyond what users would reasonably expect from the advertised functionality.

Context-Inappropriate Capability

Medium

• Confidence: 87% confidence
• Finding: The init flow opens a browser and appends tracking-style query parameters ('from=cli-skill' and skill name) while onboarding users to obtain an API key. For a synopsis-oriented skill this is unrelated behavior that can leak usage attribution data and normalize credential collection flows users did not expect.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The skill enables implicit invocation without any visible activation boundaries, exclusions, or user-consent guardrails. That can cause the agent to auto-select this skill in loosely related contexts, leading to unintended prompt injection exposure, inappropriate task routing, or unexpected use of the skill when the user did not explicitly request it.

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:128