Something weird I'm playing around with: Heisentools. Borrowed from kyle kukshtel's heisenfunction idea combined with autoresearch –– basically you make a fake service and get an agent to rewrite mock dot py until it mirrors the real tool calls it has observed and can extrapolate to new ones. This lets you simulate any service and avoid rate limits, context drift, and polluting your database with test calls when testing agents. I tested it on real slack messages but it theoretically works…