KCP Simulators¶
March 9, 2026 · LinkedIn
6 reactions · 0 comments · 654 views
We released five Java simulators for the Knowledge Context Protocol. Why? Because spec writing without runnable tests is incomplete.
scenario3 shows a RogueAgent exploiting four attack vectors against a 5-agent system. scenario4 runs two agents with identical tasks but different ethics: PoliteAgent respects advisory rate_limits. GreedyAgent ignores them and wins. Both logged.
This is where design gets honest. You learn what your spec actually enforces vs what it asks agents to choose. KCP's rate_limits don't stop a determined agent — they make violations visible.
Most protocols hide this distinction in prose. We wanted it runnable, testable, debatable. The simulators aren't proof the system is secure. They're proof the system is auditable.
- https://lnkd.in/e94JbT8e