I have built SeenRelay because I noticed that AI agents may repeat paid or slow read-only verifications. I didn't want to simply build another 'cache', but a system that first tries to see if it is indeed useful and needed, before actually being implemented, so I made available the command: npx seenrelay scan . It runs locally, doesn't upload code, doesn't modify your project and doesn't activate reuse. It can output only three results: USE / DO NOT USE / INSUFFICIENT EVIDENCE. USE — the measured workload shows that SeenRelay can add value after stronger local or provider-native controls have been considered. DO NOT USE — a simpler or stronger native approach is better, so SeenRelay should stay out of that workload. INSUFFICIENT EVIDENCE — there is not enough real workload data yet to justify either conclusion. DO NOT USE will appear if ETag or cache or provider-native controls are better, so seenrelay will stand aside. SeenRelay is integrated into Claude Code repository-hosted pending Anthropic review, so not yet approved. Feed-back and real repo testing would be highly appreciated instead of upvotes.