Merge and deploy the Handoff "In progress" page (work in progress with conversation links)

Written 2026-10-05 ~14:30 UTC. RE-VERIFY: everything below is a write-time snapshot. Re-check each PR with gh pr view <n> --repo CodexCoder21Organization/<repo> --json state,mergeStateStatus,mergedAt before acting.

Mission

User request (verbatim): "We should update https://www.handoff.wasmserver.com/ to have a page showing work in progress (active/claimed handoffs / active conversations), and include links to the relevant/associated aiclisupervisor conversation like https://aiclisupervisor.wasmserver.com/conversation?supervisor=sup-e6b06a2bfb08&id=conv-c05a01dd168cb764d9a26cbfbf18ab022bfb8f1b6394fad3e8cbc7c9f6688045"

The user chose:

• the full chain: a manager catalog sessionIds filter → a Handoff service projection → a WUI page;
• "deploy after merge": deploy only once the user has approved the merges.

What was built

• AiCliSupervisorManagerEmbedded: listConversationsJson accepts sessionIds (1–200 non-blank strings) and returns only rows with those harness-native session ids. This is version 0.0.62, not yet published. The API PR documents the field.
• HandoffApi 0.0.21 (published): adds WorkInProgressProjection / WorkInProgressItem / WorkInProgressClaim / ConversationTranscriptRecord, and HandoffService.getWorkInProgressJson(). The latter is a default method that throws UnsupportedOperationException on older services.
• HandoffEmbedded 0.0.27 (published): getWorkInProgressJson() lists open handoffs that have claims or an ACTIVE attached conversation. It resolves each claim's ConversationReference against the manager catalog as follows:
  • One batched read: chunks of 200, at most 4 pages per chunk.
  • Time-bounded by supervisorSnapshotTimeoutMs, on a 2-thread pool; when that pool is full, the call answers "unavailable" immediately rather than queueing.
  • Answers are cached for 15 seconds.
  • A row links only when both harness and sessionId match exactly. This guards against older managers that ignore the filter.
  • When the catalog is unavailable or incomplete, the projection carries the reason, and the page shows it as a warning.
  • The constructor signature is unchanged on purpose: Java cold-start launchers call the 7-argument constructor.
• HandoffServiceServer 0.0.26: adds the getWorkInProgress RPC and the sandbox client method.
• HandoffWui 0.0.26: the /in-progress page.
  • It is linked from the top of the sidebar.
  • Cards are ordered by recent activity.
  • Each claim links to the "Live terminal ↗" (/session) and/or the "Transcript ↗" (/conversation, when indexed).
  • A "Claims left open" section holds stale claims.
  • There is an empty state, and a notice when transcript lookup is unavailable or not configured.
  • Screenshots are in the PR description.

Relevant PRs / refs

All five Handoff-side PRs are green on required checks and have passed my final review. They are not merged, because merging waits for explicit user approval (Opus-mode rule).

Notes on the WUI PR's non-required check, kotlin.build (kompile-remote-build):

• It failed without running any test. Run dae54786 hit "Provisioning deadline exceeded … UrlResolver is closed; cannot fetch sandbox bytecode for 'url://digitalocean-droplets/'".
• The same check is green on main.
• Only that check run was re-requested (check-run 111778015219), so the green required check in the same suite was not wiped.

Notes on the manager embedded PR 32:

• Sibling https://github.com/CodexCoder21Organization/AiCliSupervisorManagerEmbedded/pull/30 (0.0.61, exact-id search) is already published and deployed, but not merged.
• 0.0.62 must contain PR 30's change. So: wait for PR 30 to land, rebase PR 32 (the stacked branch already shows the resolution), publish 0.0.62, then bump the embedded pin in AiCliSupervisorManagerServiceServer. That repo currently pins embedded 0.0.61 at server version 0.0.44, and its PR 36 is open.
• PR 32's red CI comes from four heavy tests (process-halt, large-state, notes convergence) that also fail on unmodified main today.

Deployed or published but not merged:

• Published to kotlin.directory: handoff:api:0.0.21 and handoff:embedded:0.0.27. Both are built from the branch heads above and are not yet on main.
• Nothing in this effort has been deployed to production. HandoffServiceServer and HandoffWui are still running their previous versions.

Next steps

1. Get user approval to merge the five Handoff-side PRs. Then merge in this order: HandoffApi 17 → HandoffEmbedded 20 → HandoffServiceServer 28 → HandoffWui 32, plus AiCliSupervisorManagerApi 17. These repos use merge queues (GraphQL enqueuePullRequest). Never use --delete-branch.
2. Deploy HandoffServiceServer first, then HandoffWui (container-nursery-deploy skill; see the memory "Handoff ContainerNursery deploy recipe"). Then verify https://www.handoff.wasmserver.com/in-progress live and check that the first frame has no error banners.
3. Without the manager redeploy, the page still works but degrades: the manager ignores sessionIds, so only exact matches on the catalog's newest page get links, and the page shows the "does not support the sessionIds filter yet" warning.
4. Restarting the manager causes an ~80-minute conversation-search outage (a 2 GB SearchableBucket replay holds the search lock), so the user must choose when it happens. Once they agree: after PR 30 lands, rebase PR 32, publish 0.0.62, bump the ServiceServer pin, then deploy.

Operational knowledge

• gh pr edit silently fails on these repos; use gh api -X PATCH repos/<o>/<r>/pulls/<n> -f body=....
• The url://buildtest/ remote test service was flaky today: persistent RPC connections closed mid-request. This was recorded as a challenge. Retry on transport errors only; use --local for small targeted runs.
• The /code box has only ~1 GB free; parallel local test runs got OOM-killed.