v0.126.0

What's new

Operators can now see why an active gateway run is still draining, rather than seeing only a generic running state. The live status reports one of four bounded reasons—agent work, a pending reply, background-task completion, or completion confirmation—and refreshes only when the reason changes; approval and question prompts take precedence. The reason is in-memory status only, is omitted from run summaries, and is cleared when the event loop exits. #1954

Gateway run records are now more likely to reach an accurate terminal state when storage writes fail. Terminal writes retry with bounded attempts and re-check record ownership before writing, while failure replies and operator updates proceed without waiting for the write; a periodic sweep reconciles stale executing or acknowledged records during gateway uptime, leaving queued work untouched. #1950

Full changelog

0.126.0 (2026-10-27)

Features

• gateway: show why a draining run has not finished in its operator status (#1954) (df75304)

Build System

• deps: update mitmproxy/mitmproxy Docker tag to v12.3.1 (#1944) (32f497c)
• deps: update Systematic to v3.26.3 (#1952) (5ce20ab)
• dev: update dependency tsdown to v0.23.2 (#1949) (4636996)

Bug Fixes

• deps: update dependency @aws-sdk/client-s3 to v3.1156.0 (#1940) (7f4d231)
• deps: update dependency effect to v3.22.4 (#1945) (86328ca)
• gateway: retry terminal run writes and sweep stranded runs while the gateway runs (#1950) (f9262c5)

Miscellaneous Chores

• dev: update dependency @vitest/eslint-plugin to v1.6.28 (#1942) (50ead05)