Naomi Hoogeweij

Evidence-based interaction friction detection and pre-generation constraint enforcement for OpenClaw agents. What it does Most agent interaction failures are not intelligence failures — they are relational mismatches. A technically correct response can still create friction through cliché empathy, unsolicited advice, helpdesk tone, excessive structure, or repetition. Friction-guard learns these patterns per user and injects constraints into the model's system context before generation, so the model produces better output instead of having bad output filtered after the fact.