skills.rest
maruakshay
maruakshayCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

multimodal-security

Assess multimodal systems for OCR prompt injection and vision-language vulnerabilities.

4|Updated Apr 27, 2026
Search Tags:#ocr#security#validation#multimodal#vision-language#prompt-injection
One-click install
npx skills add https://github.com/maruakshay/mii-ai-security --skill multimodal-security
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: multimodal-security
Source: https://github.com/maruakshay/mii-ai-security/tree/main/skills/multimodal-security
Command: npx skills add https://github.com/maruakshay/mii-ai-security --skill multimodal-security

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Review a multimodal AI system for adversarial images, OCR prompt injection, hidden text, typographic attacks, unsafe visual grounding, and vision-to-action trust-boundary failures. This guide helps security engineers map attack surfaces, validate inputs, and implement guardrails around vision-derived content.

Core Features & Use Cases

  • First Principle, Attack Mental Model, Control Lens, and containment patterns for vision-language systems.
  • Evaluate vulnerabilities like hidden text injection, typographic attacks, adversarial overlays, and screenshot-as-instruction, then apply deterministic validation and policy gates.
  • Use Case: assess a product's image and OCR workflows to ensure image-derived content cannot authorize actions or modify memory without explicit non-visual validation.

Quick Start

Review the multimodal security controls and list actionable recommendations to harden a given pipeline.

Frequently Asked Questions about multimodal-security

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
What is OCR prompt injection in multimodal vision-language systems?▼

OCR prompt injection is an adversarial attack where hidden text in images manipulates multimodal systems. This skill assesses vision-language vulnerabilities like typographic attacks and hidden text injection to implement containment and validation controls.

How do I secure a multimodal AI system against adversarial images and hidden text?▼

Secure multimodal AI systems by mapping attack surfaces and applying deterministic validation with policy gates. This skill guides input validation, provenance tracking, and safe decision-making to prevent image-derived content from authorizing unauthorized actions.

How can I prevent vision-derived content from modifying memory without explicit validation?▼

Prevent unauthorized memory modification by implementing non-visual validation and policy gates. This skill evaluates vision-to-action trust boundaries to ensure image-derived content cannot bypass explicit security controls or alter system state.

What controls are needed for typographic attacks and screenshot-as-instruction vulnerabilities?▼

Controls for typographic attacks require deterministic validation and containment patterns at the vision boundary. This skill specifies technical requirements for input validation and provenance tracking to mitigate screenshot-as-instruction vulnerabilities.

When do I need provenance tracking for image and OCR workflows?▼

Provenance tracking is needed when deploying vision-based features that process untrusted image inputs. This skill helps security engineers implement tracking and guardrails to ensure image-derived content flows through validated trust boundaries.

Can this multimodal security assessment guide incident response and design reviews?▼

This multimodal security assessment guides incident response and design reviews by detailing validation and containment patterns. It provides actionable recommendations to harden pipelines against vision-language vulnerabilities and OCR prompt injection.

Related Skills

Explore More Software Engineering Skills→
jassicsjassics
6

multimodal-security-review

Review multimodal AI application input handling for injection and parsing security risks.

Community
Advanced
pluginagentmarketplacepluginagentmarketplace
1

smart-contract-security

Audit Solidity smart contracts for vulnerabilities and generate prioritized security assessments.

Community
Advanced
uphiagouphiago
1.1k

hunt-llm-ai

Identify exploitable LLM/AI vulnerabilities crossing trust boundaries with validation gates.

Community
Advanced
jassicsjassics
6

multimodal-injection-test

Test multimodal AI systems for cross-modal prompt injection vulnerabilities.

Community
Intermediate