# Select, execute and retry tests

Inspect a catalog before choosing a subset. Selection uses resolved test IDs and inherited metadata; recognized definitions and needed inherited settings validate only when execution reaches a selected test. Unselectable files report uncertain coverage; invalid reached definitions are recorded/skipped while independent tests continue.

## Choose a subset

```sh
node tooling/testing/plan.mts inspect --catalog <project>/tests
node tooling/testing/plan.mts start --catalog <project>/tests --output <new-run-directory> --runtime <project>/testing-runtime.json --tag smoke
```

Without tag/ID selection, all tests run. Use qualified `--ids CATALOG:SECTION/ID,...` to name exact scenarios. The [first-run guide](../getting-started/first-test-run.md) selects one maintained fixed-save test, avoiding the deliberately broader sample smoke selection.

Keep `start` alive while the coordinator services requests. Tests start in filename traversal order. The run uses fresh agents and isolated browser contexts; fixture cleanup remains necessary even with isolation.

## Retain partial work

The checkpoint is the authoritative snapshot. Completed progress survives execution failure. Keep output and `raw-evidence/` private until policy review produces a separate export. Do not treat missing screenshots as a passing attempt merely because an agent described the expected behavior.

Cancellation records a marker; the coordinator must stop/finish the worker and service cleanup requests. An interrupted coordinator can reconnect to pending work in the still-running process. Process-restart resumption of in-memory fixtures is not promised.

## Retry explicitly

```sh
node tooling/testing/plan.mts start --catalog <project>/tests --output <existing-run-directory> --runtime <project>/testing-runtime.json --retry --ids catalog:profile/profile-save-roundtrip
```

Earlier attempts remain. The resolved execution revision must match, and fixtures are prepared again. Default `any-failure` summary keeps earlier failures visible; use an intentionally selected policy when integrating modules directly. Changed definitions or legacy revisions require a new run/directory.

## Review

Use [review and promotion](review-results-and-create-bug-reports.md) to inspect all attempts and produce a bug report. [Runner reference](../reference/runner-commands.md) and [canonical execution guidance](../../../integration/agent-testing.md#completion-retries-and-reporting) contain command and checkpoint detail.

With several applications, the coordinator asks the human which application even when a default exists. Within that application include all configured catalogs unless explicitly narrowed. Case-insensitive tags support all/any/exclude. A no-match selection executes nothing and exposes alternatives and unselectable coverage gaps; changing the scope requires a new explicit selection.