Restore TinySite required CI checks, then review the three PRs and assess the earlier home-page delay

RE-VERIFY: snapshot updated 2026-10-08 09:01 UTC. Re-check all PRs with gh pr view <full URL> --json state,headRefOid,mergeStateStatus,statusCheckRollup; verify branches using git ls-remote. This sweep worker never merges, closes PRs, completes the handoff, or deploys; the orchestrator decides those actions.

Mission

The owner asked for TinySite, a small website hosting service with nested assets and expiration, then a CLI that always connects over url:// and shared rules with a README pointer to https://github.com/CodexCoder21Organization/PlanRepository/blob/main/workstreams/SimpleFileSystem.md. This sweep follows the user's request to finish useful open handoffs, skip deployments and disk-exhaustion work, and leave final decisions/reviews to the orchestrator.

Completed engineering work in this sweep

1. Verified all three original PRs are still open; main in each repository remains the skeleton. All three branches were already current with main. None of the original engineering work is superseded.
2. Integrated published tinysite.conventions:tinysite-conventions:0.0.1 into the WUI's existing initial PR. Removed the copied SitePaths implementation and retained small adapters for its published names and exact HTTP refusal wording. Added the WUI README's requested stopgap/SimpleFileSystem note and corrected the CLI README's obsolete own-copy statement. Conventions README now correctly describes the coordinated branch implementations.
3. WUI local baseline path/id scenario passed 1/1. After the migration, that scenario passed 1/1. The first full local suite passed 11/13; both browser failures were missing system libraries. The screenshot skill's unprivileged setup installed working ARM libraries, but the unchanged browser scenario still failed: the tests hardcoded x86 library paths. Corrected both scenarios to derive the triplet from os.arch and removed their read of the process LD_LIBRARY_PATH. Each fixed browser scenario passed 1/1 and the final full WUI suite passed 13/13. No assertion, timeout, or iteration count changed.
4. CLI full local suite passed 9/9. These runs start real isolated storage/application instances; no production-backed test was used.
5. Pushed all code/documentation changes and verified remote SHAs. Updated all three PR descriptions with sweep provenance and this handoff URL. No source change exists only locally. No production change or new Maven publish occurred.

Relevant PRs / refs

| TinySiteWui evidence only | https://github.com/CodexCoder21Organization/TinySiteWui/tree/sweep/w6-tinysite-evidence-20261008 | 6f39d958658209fdcf840ef7ae4e1deb88e0ff43 | no PR; records only | Diagnostic logs, local reproducer/result summaries and full worker findings | Pushed, evidence only; not product work to merge |

Current CI evidence

• Original CLI head e2fa8f2 had a red required check whose entire raw log was [2026-10-07 04:11:36] Build was outpaced: remote submission was interrupted before its run ID was recorded. It did not contain a test failure or stack trace.
• The old all-admission-stopped claim is stale: today's recent build list includes actively testing/building runs. However, two remote verification commands in this sweep connected and passed health, then failed before tests with [FAILED] RPC request 'getBuildRun' to service 'buildtest' timed out after 30000ms. The remote service may be slow or unresponsive. Local mode was used after those failures.
• New CLI suite 102273753380 remained queued with zero dispatched runs. Watchman sampled the same app's main suite 101598245431, completed with two runs, which distinguishes delivery loss from a deliberately idle app. Three receiver warm probes answered 404 promptly; issued exactly one gh api -X POST repos/CodexCoder21Organization/TinySiteCli/check-suites/102273753380/rerequest at 08:47 UTC. Verify a new check/run ID before treating it as recovered; do not blindly rerequest again.
• Read the WUI host build log through HCF: run 93a3b963 finished at 08:48:51 with Build+test complete: 13 passed, 0 failed, 13 total and status COMPLETED. GitHub still shows IN_PROGRESS on the exact final head. Do not rerequest/delete this successful run to make reporting appear fixed.
• Shared CI process PID 3784621 started 08:47:59 from /root/ContainerNursery-uploads/jars/kotlin-build-ci-84213905-20260929.jar. Its stdout restores 44 workers and 693 records, then reports failed cancellation recovery against already-COMPLETED unrelated runs and a finalizeUpload failure while the BuildTest connection was not established. There is no TinySite entry in that process stdout. The underlying restart/reporting/CLI-dispatch mechanism is not yet proved; these are shared infrastructure observations.
• Both build-watchman 0.0.21 instances were explicitly stopped and exited 130 after capturing these blockers. No background task remains. The last GitHub queries still show WUI IN_PROGRESS and CLI suite total_count=0; conventions SUCCESS/CLEAN.

Earlier home-page delay — unresolved, not fixed

Live /health answered 200 (0.134 seconds warm) and home answered 200 (4.929 seconds initially, 1.798 seconds later). After the configured 300-second idle period, a normal read of /health took 9.215 seconds and HCF showed a fresh WUI JVM PID 3806218 started at 08:52:57. The earlier missing active-container/process entry was lazy idle behavior, not proof of lost process bookkeeping. Startup stdout contains UrlResolver initialization and server-start messages but no per-phase timing. Cold startup is a measurable contributor; it does not explain the historical warm-home 17–20+ second delay. HomeServlet makes one listDrops call, with no per-site reads. No reliable bug reproducer was obtained and no latency fix was guessed.

HCF access works with ~/.config/hardware-control-fabric/client_cert.pem, fabric_key.pkcs8 and server_cert.pem. The skill's default fabric-client-cert.pem filename was stale. Use coursier launch community.kotlin.hardwarecontrolfabric:cli:0.2.6 -r https://kotlin.directory -- -H 198.199.106.165 -c <client_cert.pem> -k <fabric_key.pkcs8> --ca-cert <server_cert.pem> then read-file. Relevant remote paths: /root/buildtest-data/runs/93a3b963/build.log, /root/ContainerNursery/data/container-logs/kotlin-build-ci-84213905-20260929-0822c43c.stdout.log and .stderr.log, /root/ContainerNursery/data/container-logs/tinysite-wui-64107721.stdout.log. Full snapshots are in the evidence branch above under handoffs/artifacts/2026-10-08-w6-tinysite; no key/certificate material is included.

Deployed/published state

The original handoff reported the live JAR was built from dcae604 on tinysite-initial, before later test changes. HCF sha256sum independently confirms /root/ContainerNursery/apps/tinysite-wui.jar has SHA256 e24bd6161d4ebf11436e87d0e59e0e4cfe152d491e0357c8961ed2362b3d160a, matching that historical prefix; the source commit provenance remains reported rather than proven by an embedded commit ID. The route still points to /root/ContainerNursery/apps/tinysite-wui.jar using danger_jarfile and declares url://filedrop/ as a dependency. The new shared-conventions source has not been deployed. Conventions 0.0.1 is already published and was exercised by the local WUI suite; its source remains in the open conventions PR. Do not overwrite that claimed version.

Next steps

1. Restore shared CI reporting for WUI completed run 93a3b963 without rerunning it. Diagnose CLI suite 102273753380 delivery/dispatch after its single warmed rerequest before taking another recovery action. Keep kotlin.build (remote) required on the exact final heads even if another check succeeds.
2. Confirm the two independent review passes against final green heads. Candidate-head reviews found no blocking issue in the rules migration, architecture harness correction, or CLI README change; conventions final-head reviews are already clean.
3. Orchestrator decides whether to merge the three coordinated PRs. No further non-deploy pin bump or shared-rules follow-up is needed after they land.
4. For the still-unexplained historical home-page delay, use the working HCF access and profile timing around the live WUI listDrops call before choosing a fix. Do not call this resolved merely because later requests were faster.
5. All deployment/restart/route updates are skipped by the sweep. Do not complete the handoff in the worker; report verdict and release its claim for orchestrator review.

Operational notes

Use build-watchman plain --pr mode; never --to-merged or --until-merged in this worker. The installed gh pr edit fails on the deprecated GraphQL projectCards field; updating descriptions via gh api PATCH repos/<owner>/<repo>/pulls/<id> with a JSON body file works. Chromium libraries can be installed without root by ~/.codex/skills/webapp-screenshots/scripts/setup-chromium.bash. Browser cache/library setup is outside the source repositories. The report-challenge CLI automatically enqueues/merges, so it was not run under the sweep's no-enqueue/no-merge rule; this handoff preserves the infrastructure/tool friction.

Sweep verdict

BLOCKED on required CI dispatch/reporting. Conventions alone is green and reviewed; no PR should be closed as obsolete. WUI/CLI candidate-head reviews are clean, but cannot substitute for missing required checks. All code and necessary diagnostic records are remote, no deployment/Maven publish happened in this sweep, and the worker claim is released after its BLOCKED report. Orchestrator owns merges and completion.