prompt-injection-red-teaming

What problem does it solve?

Red-teaming an LLM-based system is often ad-hoc and brittle. This Skill provides a structured methodology to identify, categorize, and measure prompt-injection vulnerabilities so defenses can be validated and improved over time.

Core Features & Use Cases

• Systematic attack taxonomy across five classes (direct overrides, indirect injections, obfuscation, multi-turn manipulation, and role/context manipulation).
• Probing library design with documented payloads, expected outcomes, and quantitative bypass metrics.
• Continuous evaluation and governance workflow, including regression suites and deployment gating.
• Use Cases: security teams validating defenses in development, production readiness assessments, and ongoing security assurance for prompts and tools.

Quick Start

Run a red-team evaluation against your model using the defined probe library and document the results.