v1.23.44

• Phoenix ID's default host is now id.phoenix.so (PHNX-4424). agents auth login, whoami and logout, and a session-backup Worker set up from now on, talk to https://id.phoenix.so unless PHOENIX_ID_BASE says otherwise; a Worker set up earlier keeps checking id.byphoenix.com, which still serves. id.byphoenix.com keeps serving, and a session signed in through either host is valid at both, so existing sign-ins keep working. Source: cli/src/lib/identity/client.ts.

• Far devices are no longer marked unreachable. The device probe opened a new SSH connection every time, and a cold handshake to a peer 300 ms away takes 3.4–3.8 s, past its 2.5 s budget. agents devices status, AGI EXT's device list and --device auto therefore treated such boxes as unreachable and placed runs on the launching machine. The load probe now reuses the shared SSH connection other fleet commands keep open, which answers in 0.6–0.9 s, and a timed-out probe warms that connection in the background for the next one. The budgets are unchanged. Source: cli/src/lib/devices/health.ts, cli/src/lib/devices/connect.ts.

• agents projects prs list <project> returns in about half the time on a repository with many open PRs. GitHub works out each PR's merge state (mergeStateStatus) when asked, so the one paginated GraphQL list of 100 PRs a page spent about 3 s per page on that field alone, one page after another. The list now asks only for the open PR numbers, then reads the PRs 10 to a query, at most 12 queries at a time across every repository the process reads (each costs 1 GraphQL point). On phnx-labs/agents (115 open PRs) the command went from 5.9 s to 3.0 s (median of 5 runs each), and the JSON is byte-identical. Source: cli/src/lib/github/project-prs.ts (listOpenPrs).