Back to skill

Security audit

Nano Banana Pro 图片工作室|AI-HIVE

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real AI-HIVE image-generation skill, but it needs review because its API client can send the user's AI-HIVE key to a configurable remote host.

Use this only if you intend to connect to AI-HIVE and run image-generation workflows. Do not set --base-url, AI_HIVE_BASE_URL, or a custom base_url in ~/.ai-hive/config.json unless you fully trust that endpoint; use a dedicated limited API key, confirm prompts and routing before paid jobs, and upload only media you are authorized to share.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/imagegen.py:92
Finding

API Credential Can Be Transmitted to an Arbitrary Endpoint

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
scripts/imagegen.py:14
Finding

Unpinned Runtime Dependency Installation

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (11)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The declared description promises a broad image-studio capability centered on actual media generation/editing and AI-HIVE service orchestration. The supplied code only creates a local JSON blueprint containing metadata such as project, audience, goal, platform, and format. This is not just an implementation subset; it reflects a materially different primary purpose: brief/spec generation rather than image production or editing. While the produced artifacts mention prompts and delivery checks, the code does not actually create those beyond listing them as static strings. Therefore the description significantly overstates what the code chunk does.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
93% confidence
Finding

The skill includes runnable code and operational guidance that imply access to environment variables, local files, shell commands, and network endpoints, but it does not declare any explicit tool scope or permission boundaries. This is dangerous because an agent may invoke broader capabilities than users expect, including reading local reference assets, using API keys from the environment, and making external calls to a third-party service without clear least-privilege constraints.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger phrases are broad enough to capture many ordinary image-related requests, increasing the chance that the skill is invoked when the user did not intend to authorize external generation, asset upload, or potentially billable API usage. In this context, overbroad activation is risky because the skill handles third-party uploads, routing decisions, and asynchronous task execution, which can expose user data or incur cost if engaged unnecessarily.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The default prompt auto-frames the skill as a general entry point for turning requests into a workflow, runnable AI-HIVE commands, and deliverables, but it does not constrain when that behavior is appropriate. Because the skill handles external-image generation/editing operations and production-ready outputs, broad activation increases the chance of unintended invocation on ambiguous requests, causing overreach, unwanted tool use, or unsafe content transformation without explicit user intent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The default prompt explicitly instructs the skill to produce a "production-ready Chinese workflow," which imposes a specific language choice without offering the user an opt-in or alternative. This is a natural-language locale policy concern because the skill enforces Chinese regardless of user preference.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Enabling implicit invocation without tight activation constraints allows the skill to be selected automatically based on loose semantic similarity rather than explicit user consent. In a skill that can generate runnable commands, interact with external APIs, and process marketing/e-commerce imagery, this raises the risk of accidental activation, unexpected external actions, and policy-sensitive image manipulation workflows being initiated in the wrong context.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The file exposes broader capabilities than the declared image-studio skill, including generic chat and video generation. In a skill-based environment, this scope mismatch can bypass least-privilege expectations and allow callers to invoke unintended paid or sensitive operations through a skill that appears narrowly scoped.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The module presents itself as a general-purpose AI tool rather than a specialized image-studio skill, which creates a documentation-to-behavior mismatch. That mismatch increases the chance that reviewers, orchestrators, or users grant trust or access under false assumptions, enabling misuse of hidden broader functionality.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This code file contains natural-language descriptions, help text, and runtime prompts entirely in Chinese, which effectively forces a specific language/locale on users. The file does not offer any user opt-in, fallback, or locale selection mechanism, so it conflicts with the policy against imposing a language without choice.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill can call a user-info endpoint that returns account and wallet details unrelated to image generation. This expands data access beyond business need and can expose sensitive billing or account metadata if the skill is invoked in contexts where only media operations were expected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

This code embeds Chinese-language naming in the module docstring, argument parser description, and generated JSON skill field, with no option for the user to select a language or locale. The stated policy for this review flags language or locale constraints when they are imposed without user choice or clear justification.

Content

No source excerpt is available for this finding.