Testing Agent-Written Specs by Replaying Historical CVEs Against Them

jdpressman proposes a workflow where an agent writes its own software spec, then checks it against every historical CVE to see if it would prevent each vulnerability, followed by adversarial attack testing; discussants note that even lazy specs deliver outsized returns.

2026-10-24 ~ 2026-10-24 · 4 related posts

• Testing agent-written specs against every historical CVE, then adversarial attack — jd_pressman · 2026-10-24
• Agent-driven spec validation: replay every historical CVE, then adversarially attack your own security spec — jd_pressman · 2026-10-24
• Even the laziest AI coding specs deliver outsized bang for buck — dlbydq · 2026-10-24
• Agent-designed specs tested against every historical CVE, then adversarially attacked — jd_pressman · 2026-10-24