Reticle brings Postman-style agent debugging

Hot take: this is the direction agent tooling needs to go, because ad-hoc logs and one-off playground runs break down fast once tools and multi-step decisions enter production workflows.

• –Scenario-based runs plus side-by-side model testing make prompt iterations less guessy and more reproducible.
• –Built-in eval datasets are a strong signal that the project is thinking beyond demos toward regression safety.
• –Local SQLite storage is a practical trust win for teams worried about leaking prompts, API keys, or proprietary traces.
• –The gap to watch is ecosystem integration (LangGraph, LlamaIndex, OpenTelemetry imports), which often decides whether debuggers stick.