Agent proxy behavior verifier

Detect when your proxy silently downgrades system prompts to user messages

Problem

A developer dug into proxy source and found nearly all relays move original system instructions into the user role for non-Claude-Code agents, breaking system priority and long-conversation behavior. Users only feel “something’s off” and can’t pinpoint it.

Solution

A diagnostic tool that sends a set of standard probes to a configured relay endpoint, uses known system-following tests and role-order analysis to detect whether system/user roles are rewritten or disguised, and outputs a visual report with risk scores and a direct-connection baseline.

Users

Developers running self-hosted relays with pi/opencode/Codex and relay operators

MVP

Enter endpoint and key; run system-following probes; detect role-mapping rewrites; output issue list and risk level; save historical comparisons

Revenue

Free basic checks plus one-time or subscription Pro (multi-endpoint, scheduled scans, CI integration)

Why now

Relays are everywhere and inconsistent, users are baffled by odd behavior, and the probe logic is buildable in about two weeks