upgrade-engines
Read docs/ai-engines.md and the affected driver in packages/harness-*.
Update one engine per change:
pnpm engines
pnpm engines update <kind>
pnpm engines verify
pnpm check:changed --base origin/main
pnpm test:engines -- -t 'shipped Codex'
Replace Codex with the suite's engine label. Also run the installed <label> selection for engines declaring an installed runtime. Use dedicated spending-limited OPENAI_API_KEY/ANTHROPIC_API_KEY credentials or existing machine sign-ins; shipped qualification puts shipped bytes first on PATH. Missing sign-in or quota blocks qualification. Inspect emitted JSON artifacts under OPENSTEER_ENGINE_E2E_OUTPUT or the printed directory. Do not call skipped qualification passed.
Antigravity is downloaded rather than shipped or found on PATH. A pin bump must update and re-verify every platform's archive size, SHA-256 and complete member manifest in its driver, then run the downloaded Antigravity qualification against a temporary Runner runtime cache.
For a new model, first check the account's real engine catalog. Update its engine when required; never hardcode a model the engine cannot describe. Check native ID routing, context/output limits and options in affected driver tests. Review protocol and sandbox changes, including Codex's approved feature list. Reconnect the Runner to refresh accounts after release.
For a running local oMLX server, qualify a model that fits that computer: