v0.13.0

What's Changed

• feat(summarizer): predictable model selection, drop cheapest-model oracle by @lvndry in #270
• fix(input-service): stale cleanup no longer removes replacement handler by @lvndry in #274
• refactor(prompts): drop default-prompt fallback, fail if persona unresolvable by @lvndry in #271
• feat(llm): capture and persist provider reasoning parts in history by @lvndry in #275
• feat(ui): recall queued messages into the input with arrow-up by @lvndry in #273
• feat(llm): replay persisted reasoning parts across tool loops by @lvndry in #276
• feat(agent): decompose loop behind observer seam + reframe default persona by @lvndry in #272
• feat(run): add --conversation flag for per-chat memory in jazz run by @lvndry in #280
• feat(ui): UI/UX overhaul phases 1-2 — default polish and reasoning visibility by @lvndry in #278
• refactor(config): separate runtime UI state from user config by @lvndry in #283
• feat(cli): confirm before agent delete, add --yes/--force by @lvndry in #282
• feat(evals): everyday-assistant eval harness (#0) — runs, guardrailed, model-discriminating by @lvndry in #277
• feat(ui): UI/UX overhaul phases 3-5 — theme unification, input power, command surface by @lvndry in #279

Full Changelog: v0.12.9...v0.13.0