veld-launch-feedback-loop

# Launch the Veld Feedback Loop Veld shows the human an in-browser overlay to comment on elements, pages, and screenshots. Those comments land on a **linear queue**. Your job: drain it — pull the next item, fix it, reply, repeat — until the human clicks **Done**. The queue is stateless from your side: there is **no cursor to track**. You call `veld feedback next`, work the item it returns, then `reply`/`resolve`. That item drops off, and the next call returns the following one. ## On invocation — start looping now 1. **Find the run.** `veld feedback next` auto-selects the current run (a single active or recently-used run); if several exist, pass `--name <run>`. If there is no run at all, ask the human to `veld start <node> --name <run>` first. 2. **Enter the loop below and keep running it until you get `"ended"`.** Don't stop after one item — this is a Ralph loop; you keep pulling. ``` loop: out = veld feedback next --wait --name <run> --json → result "item" : work it (below), then reply (or resolve) → result "timeout" : call next again (nothing waiting yet) → result "ended" : the reviewer clicked "Done" → stop the loop ``` `next --wait` blocks in the CLI (polling ~1×/sec, up to ~4 min) then returns `timeout`. On `timeout`, **just call it again** — re-invoking is free (no cursor) and resumes cleanly even if you were killed or the session restarted: a day-old comment is still at the head next time you call. ## Working an item The `item` payload ha...