PR 28533 runtime verification
main: 37d7615a91
PR: 56dba87bc30f7497ecdf1fb9d27183c349bf35bf
Linux; repo Playwright 1.60.0; /usr/bin/google-chrome; real Codex gpt-6.1-sol via local proxy.
One dev server at a time. Separate real T3 peer on port 17777.
Project /home/user/peer-verification-repo, link, and turns created through the app.
Peer suspended with SIGSTOP to its captured process group; resumed with SIGCONT. No response mocks or database fixtures.

Web client WebSocket requests (sent-to-response; single observations, includes client scheduling):
main: peerLinks.list 2647 ms; {"_tag":"Success","value":{"links":[{"environmentId":"337382b4-07aa-4f53-b9ed-6c28072692d1","label":"box-node-34e92b95c8ea8709","urls":["http://127.0.0.1:17777"],"access":"approval-required","linkedAt":"2026-11-01T07:55:03.738Z","expiresAt":"2026-12-01T07:55:03.738Z","lastReachedAt":"2026-11-01T07:56:17.357Z","lastError":null,"status":"unreachable"}]}}
pr: peerLinks.list 451 ms; {"_tag":"Success","value":{"links":[{"environmentId":"337382b4-07aa-4f53-b9ed-6c28072692d1","label":"box-node-34e92b95c8ea8709","urls":["http://127.0.0.1:17777"],"access":"approval-required","linkedAt":"2026-11-01T07:55:03.738Z","expiresAt":"2026-12-01T07:55:03.738Z","lastReachedAt":"2026-11-01T07:56:17.357Z","lastError":null,"status":"reachable"}]}}

Real provider MCP inventory spans (server path, not whole model-turn latency):
main: 37.836 ms; exit=Success; at=2026-11-01T07:56:14.243Z
main: 3018.738 ms; exit=Success; at=2026-11-01T07:57:10.513Z
pr: 15.961 ms; exit=Success; at=2026-11-01T07:59:02.303Z
pr: 6.342 ms; exit=Success; at=2026-11-01T07:59:30.635Z
pr: 10.663 ms; exit=Success; at=2026-11-01T08:00:34.033Z
pr: 10.992 ms; exit=Success; at=2026-11-01T08:00:41.553Z

PR authenticated forwarding spans:
PeerMcpClient.call: 3029.538 ms; exit=Failure
PeerMcpClient.call: 29.236 ms; exit=Success

Observed in the real client:
1. Main, awake peer: reachable; remote project list succeeded.
2. Main, suspended peer: inventory waited for descriptor timeout; unreachable.
3. PR, same suspended peer and stored success: immediate stored reachable.
4. PR, remote project list while suspended: did not answer; inventory unreachable with that lastError.
5. PR, awake again before retry: still stored unreachable.
6. PR, real remote project list retry: succeeded; inventory reachable, lastError=null.

Existing checks run this turn:
blacksmith-check run --id <task-owned-worker> "vp i && vp test run apps/server/src/peer"
5 test files passed; 67 tests passed; 47.09 s.
GitHub CI: applicable checks pass on exact unchanged head; no code changes or pushes.

Not exercised: native mobile/Electron clients; Tailscale/relay; non-loopback 10-second timeout; live expiry/revocation/identity replacement. Focused tests cover expiry and identity fencing.
Known existing review concern: agent/CLI descriptions still say inventory checks whether each peer answers now, while PR behavior deliberately uses recorded outcomes. No runtime failure observed from that text in these turns.