[adenzhou1350]

Summary

The agent skill recommends --quiet --output json, but non-streaming text chat currently prints only the extracted answer in that combination. This drops message metadata and usage, and ordinary text answers are no longer valid JSON.

Honor the resolved output format regardless of quiet mode, matching the existing image command behavior. Plain-text output, dry-run and streaming are unchanged. The regression now checks the complete fixture, with plain-text controls for both quiet settings.

Validation

Linux, Bun 1.4.3, task-private install using the frozen lockfile:

• Same focused tests on unchanged production: 7 pass, 1 regression failure. Patched production: 8/8 pass.
• Full bun test: 613/613 pass; TypeScript and ESLint pass.
• Two independent full-diff reviews, as requested by AGENTS.md.

Mock HTTP responses only; no live API, Windows, compiled-binary or official-CI validation is claimed. AI-assisted fix and tests.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}

[adenzhou1350]

@RyanLee-Dev could you take a look when you have a chance?

This still reproduces in installed mmx-cli 1.0.18 on Windows / Node 24.14.1: with a loopback Messages fixture, --quiet --output json exits 0 but prints only the answer, dropping usage. JSON without quiet and both text-mode controls pass. This is compiled-release baseline evidence, not fixed-binary or live-API validation. The existing branch is still mergeable.