oh-my-codex v0.8.3

Released: 2026-03-06

This is a focused hotfix release for the Gemini team-worker path shipped in the 0.8.2 dev release line.

TL;DR

• Fixes Gemini worker startup in team prompt mode by launching workers with --approval-mode yolo -i "<initial inbox prompt>" instead of depending on stdin for the first instruction (#585).
• Prevents non-Gemini default models such as gpt-5.3-codex-spark from being passed through to Gemini workers unless the configured model is explicitly a Gemini model (#585).
• Adds targeted runtime and tmux-session regression coverage for the Gemini prompt-launch path (#585).
• Includes a small test-only hardening for the notify-fallback watcher so full-suite release validation remains stable under load.

What changed

1) Gemini prompt-mode workers now start with an explicit initial prompt

Gemini workers launched through OMX team prompt mode are now started with an explicit initial instruction:

• --approval-mode yolo
• -i "Read and follow the instructions in .../inbox.md"

Why this matters:

• removes dependence on stdin-delivered bootstrap text for Gemini startup
• aligns worker bootstrap with Gemini CLI expectations in prompt mode
• fixes the broken worker bring-up path reported in the hotfix PR

2) Non-Gemini default model passthrough is filtered for Gemini workers

Gemini workers no longer inherit non-Gemini default models by accident.

Current behavior from this release:

• explicit Gemini models still pass through
• non-Gemini defaults are omitted for Gemini workers
• mixed-provider team configs avoid invalid startup argument combinations

Why this matters:

• prevents invalid provider/model pairings during worker launch
• preserves cleaner mixed-provider CLI interoperability
• reduces surprising failures in prompt-mode and mapped-worker setups

3) Regression coverage was expanded for the hotfix path

This release adds focused tests covering:

• prompt-mode Gemini startup argument construction
• runtime startup behavior for prompt-launched Gemini workers
• translation behavior when default models are non-Gemini

4) Full-suite verification was stabilized

Release validation also hardened a flaky watcher test so the full suite reliably waits for watcher readiness before asserting streaming EOF-tail behavior.

Why this matters:

• keeps release verification deterministic under heavy suite load
• preserves the intended watcher behavior instead of relying on fixed sleeps
• does not change shipped Gemini runtime behavior

Related PRs and issues

Merged PRs in this release

• #585 — fix(team): seed gemini workers with prompt-interactive launch

Scope note

• Functional release scope is centered on PR #585, the Gemini worker startup hotfix after the 0.8.2 dev release line.
• Release validation also includes a small test-only stabilization in src/hooks/__tests__/notify-fallback-watcher.test.ts so the full suite remains reliable under load.
• Final tracked change set for the release branch: package.json, package-lock.json, CHANGELOG.md, and the watcher test hardening.

Verification summary

Release verification evidence is recorded in docs/qa/release-readiness-0.8.3.md.

Planned release gates:

• npm run build
• npm test
• npm run check:no-unused
• CLI smoke checks (--help, version, status, doctor, setup --dry-run, cancel)
• Gemini-targeted regression checks from PR #585

Thanks for using oh-my-codex. If anything regresses, please open an issue with reproduction steps, logs, and your CLI/runtime details.