fissure

Automate adversarial testing of AI models across browser, Ollama, and OpenRouter backends.

2|1|Updated Apr 22, 2026

Search Tags:#model evaluation #ai security #adversarial testing #ai safety #red-teaming

Ready to run

npx skills add https://github.com/m4xx101/fissure --skill fissure

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: fissure
Source: https://github.com/m4xx101/fissure/tree/main
Command: npx skills add https://github.com/m4xx101/fissure --skill fissure

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires requests, requests-oauthlib, hermes-tools, ollama, openrouter, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill automates the testing of AI models for safety boundary robustness, allowing for efficient model evaluation and security assessment.

Core Features & Use Cases

Adversarial Testing Harness: Provides a comprehensive framework for testing AI models through various adversarial techniques.
Browser, Ollama, OpenRouter Support: Tests models across different platforms and backends, including browser-based interactions, local Ollama models, and OpenRouter cloud services.
Automated Orchestration: Employs an agent-based architecture for autonomous testing and analysis, without requiring manual Python scripting.
Use Case: Consider a scenario where you need to evaluate a new AI model for its robustness against adversarial inputs. Fissure can automatically generate and test payloads, analyze responses, and provide comprehensive results.

Quick Start

Use the fissure skill to test the AI model at the URL 'https://gandalf.lakera.ai/baseline'.

Frequently Asked Questions about fissure

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

FAQPage Schema

How do I automate AI safety boundary testing for my models?▼

AI safety boundary testing is automated by deploying an agent-based architecture that generates and tests adversarial payloads, analyzes responses, and delivers comprehensive model evaluation results without manual Python scripting.

Can I run adversarial testing on local Ollama models and OpenRouter backends?▼

Adversarial testing supports local Ollama models, OpenRouter cloud services, and browser-based interactions, allowing you to evaluate AI safety boundaries across multiple distinct platforms and execution environments.

Do I need manual Python scripting to perform red-teaming on AI systems?▼

Red-teaming AI systems requires no manual Python scripting because the agent-based architecture autonomously orchestrates adversarial testing, payload generation, and response analysis for comprehensive security assessment.

What's the best way to evaluate an AI model for robustness against adversarial inputs?▼

Evaluating AI model robustness against adversarial inputs is best handled by an automated testing harness that generates payloads, probes safety boundaries, and analyzes responses to provide comprehensive security assessment results.

What Python dependencies are required to set up an automated adversarial testing harness?▼

Setting up an automated adversarial testing harness requires Python libraries including requests, requests-oauthlib, hermes-tools, ollama, and openrouter to execute testing and analyze model evaluation responses.

Why use an autonomous agent for AI security evaluation instead of manual testing?▼

An autonomous agent for AI security evaluation eliminates manual scripting overhead, continuously generating and testing payloads to probe safety boundaries and analyze responses more efficiently than manual techniques.