Giving AI agents vague bug reports is painful

Honestly tired of getting bug reports like "the button is broken" with zero console logs, no network payload, and no context.

Now that we use Cursor and Claude Code to fix bugs, it's even worse because the models just start guessing and hallucinating without real telemetry.

We started building caes.dev to bundle the screen recording, console, and network logs in one click so you can hand the full context directly to your agent or issue tracker.

Curious how you guys deal with this right now? Are you still manually typing repro steps or do you have a better workflow?