v0.4.47

Changed

• Added --no-fallback to relay schedule. A task using this flag stays on its selected provider when it reaches a usage limit instead of trying the other provider.
• Claude Code tasks using this flag retain Relay's resume-after-reset behavior. Tasks without the flag follow the existing fallback setting.

v0.4.46

fix: Codex rough edges - cancel race and resume

- A run that finishes cleanly before a cancel's kill lands now records its real result
  (a CLI-cancelled Codex run posted the pin, then showed 'cancelled').
- Claude-transcript checks (missing-transcript fallback, collision guard, silent-resume
  retry) are skipped for Codex: they never find ~/.codex sessions, so every Codex resume
  silently became a fresh run.
Bump to 0.4.46.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MM2obvzrm9ao4EMShJjyAg

v0.4.45

feat: fall back to the other subscription when a task hits a usage limit

A run stopped on a Claude limit retries once on Codex (and vice versa) before Relay pauses
it for the cooldown. Fresh session in the same folder, prompt told to continue the partial
work. Only if both are limited does the normal resume-at-reset run (Claude's reset, also for
Codex originals). Settings toggle + per-direction model pickers; the card shows the fallback.
Bump to 0.4.45.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MM2obvzrm9ao4EMShJjyAg

v0.4.44

fix: close Claude login window when the session cookie lands

claude.ai finishes login with in-page navigation, so the did-navigate URL check never
fired and the window stayed open on claude.ai. Close on the sessionKey cookie instead,
then bring Relay back to the front. Bump to 0.4.44.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MM2obvzrm9ao4EMShJjyAg

v0.4.43

feat: add Sonnet 5.5 and GPT-6.1 Sol models

Sonnet 5.5 replaces Sonnet 5 in Current (Sonnet 5 moves to Legacy); GPT-6.1 Sol tops
the Codex group. Skill examples use Sonnet 5.5, and the Opus/Fable cap note no longer
splits the skill's model table. Bump to 0.4.43.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v0.4.42

feat: cap every Opus and Fable model at high effort

effortPolicy matches /opus|fable/ so future versions are covered; the CLI
rejects xhigh/max so scheduling agents can't set it, and the executor
clamps stored tasks. Bump to 0.4.42.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v0.4.41

feat: cap Opus 5.5 effort at high, default medium

Executor clamps every run (UI, CLI, skill, pre-existing tasks); the CLI
rejects xhigh/max for claude-opus-5-5; UI and skill only offer low-high.
Bump to 0.4.41.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v0.4.40

feat: run tasks on Codex (gpt-*) models; add Opus 5.5 + Fable 5.1

gpt-* model ids route through codex exec (effort via model_reasoning_effort,
session id read from the header, resume via codex exec resume). Codex limit
stops are not auto-resumed on the Claude reset clock. Bump to 0.4.40.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

v0.4.39

fix(scheduler): hold the queue on a usage limit instead of rewriting …

v0.4.38

feat(cli): --chrome flag so scheduled tasks can opt into browser access

Browser access was settable only in the app's task modal. Agents schedule
through the CLI and can't tick that box, so any browser task they queued ran
without Claude-in-Chrome tools and failed at step one.

chore: bump version to 0.4.38

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H1gpAFHgaJ3P5VAQcHx3bD