Skip to content

PRODUCT.md

Latest commit

 

History

History
35 lines (21 loc) · 2.25 KB

PRODUCT.md

File metadata and controls

35 lines (21 loc) · 2.25 KB
Raw

Product

Register

product

Users

Hackathon judges evaluating AgentGuard at HACKHAZARDS '26, and developers/security engineers who use it as a CI adversarial-testing tool for AI agents. Judges need a quick impressive demo that showcases depth. Developers need clarity, speed, and actionable results. Both audiences share a context: technically sophisticated, familiar with security tooling, skeptical of shallow demos.

Product Purpose

AgentGuard is an adversarial testing harness that stress-tests AI agents across 10 attack categories (jailbreak, prompt injection, context leak, etc.), visualizes failure cascades as force-directed graphs, and produces statistically-rigorous readiness scores. It works as a CLI, a web dashboard, a real-time proxy, and a GitHub Action. Success is a developer running one command and knowing whether their agent is safe to ship.

Brand Personality

Precise, confident, unpretentious. Speaks with authority earned by rigorous testing, not by posturing. Think security audit tool, not cybersecurity marketing. Tone is direct and technical — no hype, no cargo-cult "cyber" language.

Anti-references

  • Anything that looks like a generic SaaS dashboard (pastel gradients, rounded cards, heavy shadows)
  • "Cyber-security" cliches (skulls, shields, matrix-green-on-black, excessive glow effects, faux-holographic UI)
  • Over-designed dashboards where style obscures data
  • The current scanline CRT overlay (too theatrical — undermines credibility with technical users)

Design Principles

  • Data first, chrome second — every visual element earns its place by making test results clearer
  • Technical honesty — the interface looks like what it is: a serious testing tool. No decorative fluff, no gimmicks
  • Progressive disclosure — beginners see a clean demo flow, power users can drill into cascade graphs, raw logs, and Neo4j queries
  • Dark but not dramatic — dark background for reduced eye strain during long sessions, without theatrical effects that undermine trust

Accessibility & Inclusion

Latent: dark mode with sufficient contrast (#EAEAEA on #0A0A0A is ~15.5:1). Monospace font may reduce readability for non-technical users. No reduced-motion consideration yet. No color-blind testing done on the red-accented failure states.