[ac-pplx]

Summary

Every sandbox can now be found from the sandbox list by its identifier, however large the fleet.

The list loads the newest 1000 sandboxes (GET /admin/api/v1/sandboxes?limit=1000), reloads them periodically while the page is visible, and filters them in the browser. The admin list has no cursor and no search, and it orders by created_at DESC. In a fleet with more than 1000 live sandboxes, the oldest ones (often the long-running sandboxes a support case is about) were therefore not found by any search, and nothing on the page said the list was cut off.

Changes:

• A full identifier in the search box now names one sandbox and matches it whatever the other filters say. Accepted forms are the sbx_ ID, a bare or undashed UUID, and the pod name space-sbx-<hex>-g<N>. The placeholder now reads "ID, pod name or name".
• When no loaded row matches, the page fetches that sandbox with GET /admin/api/v1/sandboxes/{id} and shows it with a notice. The table shows as loading only until the first answer for that ID; later refetches, including ones that answer 404, do not bring the spinner back. The row is fetched again after each list load, so it stays current after an action, a Refresh or a poll; it stays shown while that runs, and a 404 clears it. An unknown ID shows the API's "Sandbox not found".
• The lifecycle policy dialog opens for a looked-up row as for any other.
• When the list is capped, a notice says so and explains the direct lookup. Only sandbox rows count toward the cap: the admin list appends the rows nodes report without a database row (rowless pool slots, observed-local sandboxes) after its limit sandbox rows.
• "Load newest" chooses 100, 1000 (the default) or the API maximum of 5000 rows. It is also available as ?limit=. A list larger than the default reloads less often than the default-size list (at most about every 15 seconds), and keeps its own refresh timestamp: console windows share one per view, and a default-size list open in another window would otherwise keep the larger list from ever reloading. An answer for a load size that has since changed is dropped.
• The overview's sandbox count reads 1000+ instead of a capped number, counting sandbox rows only.

Partial text still filters the loaded rows by ID fragment or name.

Server-side search, filters and cursor paging need admin API work and are not part of this change. Until then, this keeps every sandbox reachable by ID.

Stack

Stacked on #1242, which adds parseSandboxRef. Merge #1242 first. This branch is restacked on #1242's review follow-up. #1248 is stacked on this PR.

Validation

• npm ci, npm test (75 pass) and npm run build in ui/. The new unit cases in src/lib/sandbox-id.test.mjs cover exact lookup in every form, ID fragments, names, the limit clamp and the sandbox-row count.
• A Playwright loop against a local Kind stack with synthetic data, comparing main with this branch (the Kind space-ui image). It found no new console errors, page errors or failed requests.
• A real >1000-sandbox fleet was not seeded, because each sandbox is a Firecracker VM. The cap path is instead exercised with ?limit=2: only the two newest sandboxes load, and searching the pod name of an older one finds it by direct lookup.

Visual confirmation

With only the newest 2 sandboxes loaded, the pod name of an older sandbox finds it:

Video: after.

Before (main): no match	After: found by direct lookup

Severity and risk

Severity: high. Finding a sandbox by ID is the first step of most support cases, and it silently failed for any sandbox outside the newest 1000.

Risk: low.

• Partial-text search behaves as before.
• A full identifier now bypasses the template, status, node and age filters, on purpose.
• While a full identifier that matches no loaded row is in the search box, the page makes one GET /admin/api/v1/sandboxes/{id} on each list reload, never two at once for the same ID. Nothing extra is requested otherwise.
• Roll back by reverting the commit.

AI assistance: Claude Code wrote the code, the tests and this description, and ran the validation; a maintainer reviews it.

[github-actions]

Thanks for opening this pull request! A SPACE maintainer will take a look.

CONTRIBUTING.md explains how we review outside contributions.

Withdrawing this approval for now. Any feedback above still stands.

[johnlouie-pplx]

Approving at 06814fc, with one nonblocking ask. The direct lookup uses the existing authorized endpoint, and all 75 UI tests passed locally.

Could we give the lookup effect in sandboxes/page.tsx a per-request token and only let the current request update the result or clear the in-flight marker? I reproduced A → B → A with the real component and mocked responses: the first A request can finish last and overwrite the newer result, including replacing a valid row with “not found.” A regression test that completes those requests out of order would cover it. #1248 fixes the separate list-spinner issue, but not this lookup race. Worth fixing proactively here, though I wouldn't hold the merge for it.