What's Changed
- ci(release): drop deploy-soundcheck from the release (~2.5m) by @chelojimenez in #4346
- fix(connect): wait for AuthKit before claiming a handoff link by @chelojimenez in #4335
- feat(cli): six exit codes for
eval run --wait(E1) by @chelojimenez in #4342 - eval: let a structured launch refusal reach the caller instead of a 500 by @chelojimenez in #4348
- harness: Codex delivers the host's MCP servers as host-executed tools (COMP-39) by @chelojimenez in #4339
- eval: type toolCallId at the persistence boundary instead of casting it by @chelojimenez in #4351
- Evals v2 — E4a:
eval gate --baseline <runId>regression-delta gate by @chelojimenez in #4343 - D7 — metadata-attribution selection judge (SDK contract, tool-catalog capture, second pass) by @chelojimenez in #4341
- Evals v2 G4c: disclose the host axis before a run launches by @chelojimenez in #4350
- Changeset: keep #1131 off a line start (markdownlint MD018) by @chelojimenez in #4352
- E3: Minimal HTML report (decision summary + failures) by @chelojimenez in #4345
- chat-input: clear the file-error timeout on unmount by @chelojimenez in #4353
- changesets by @ignaciojimenezr in #4354
Full Changelog: v2.47.0...v2.48.0