---
name: relay-run-and-review
description: Run a saved Relay Test, repeat selected data values, inspect a failed Run, or export retained evidence for review.
---

# Run and review

1. Read `relay://guides/run`; for a failure read `relay://guides/debug`, and
   for sharing read `relay://guides/review`. Call `relay_panel` to choose an App,
   then call it with that `appMapId` to find the exact saved Test. Read
   `relay://app-maps/<appMapId>/tests/<testId>` for its steps. Use returned IDs
   in `relay_run_test`; select the intended target through `relay_connect_target`.
   A saved Test can run directly. The panel also returns text for tools-only hosts.
2. Call `relay_run_test`. Keep the returned workflow ID and version. Inspect
   through `relay_inspect_workflow` until the server reports completion or a
   concrete blocker. A queued request is not a completed Run.
3. For explicitly requested data values, use `relay_repeat_test` for one
   representative pilot. Inspect its result before `relay_continue_repeat`;
   continuation requires explicit confirmation and the latest version.
4. Inspect a failed Run with `relay_inspect_failure`. Return the failed check,
   expected versus observed state, target, and evidence. Propose a repair only
   when requested; proposals remain reviewable and retain the original attempt.
5. Export an existing Run through `relay_export_evidence`. Finish with the
   Run ID, exact result, evidence references and any missing captures or review.

Functional completion, captured screenshots, and human visual review are
separate results. A human makes review decisions; agents summarize evidence.
Keep interrupted or unknown outcomes inspectable and preserve target identity.