release: v1.2.0

changelog/1.2.0.md

@@ -0,0 +1,41 @@
+# 1.2.0 — 2026-07-01
+
+This release is about the **answers themselves** — they sound like you, you control their
+shape with one dial, and every card is now something you can act on.
+
+## Added
+
+- **Answers in your own voice.** BrainCue now answers *as you* — first person, on your behalf,
+  never "the candidate." A new naturalness pass strips the tells that make text read AI-generated
+  (corporate filler, hedging, robotic transitions), so a glance reads like something you'd actually say.
+
+- **One Answer Format dial — now with Story teller.** The old, muddy format/tone controls are
+  replaced by a single live control with four settings:
+  - **Key points** — terse, glanceable bullets to speak from.
+  - **Explanation** — a natural, flowing walk-through.
+  - **Detailed** — thorough, with a concrete example.
+  - **Story teller** — a short, vivid first-person story ("you are me telling my own story"),
+    great for behavioral answers.
+
+- **Regenerate any answer, right on its card.** The single toolbar button is gone — **every
+  answer card now has its own ↻**. Regenerate a specific question (even an older one in your
+  history — it re-expands), or re-solve a coding problem in place.
+
+- **Copy any answer.** A per-card **Copy** button drops the clean answer onto your clipboard —
+  paste a coding solution straight into the editor.
+
+- **A better coding solver.** Pick your **language** (JavaScript by default); solutions come back
+  **optimal** (with stated time/space complexity), **explained first** in plain terms, and with
+  **clear inline comments**. Coding answers stay focused on the problem — no résumé/JD mixed in.
+
+- **Pronunciation guide, on by default.** Hard words (rare, technical, foreign, names) get a
+  compact **"🗣 How to say it"** panel below the answer — part of speech, singular form, and a
+  simple respelling — while the answer text itself stays clean and natural to read aloud.
+
+## Under the hood
+
+- Every feature was designed against the real codebase and put through an adversarial,
+  multi-agent review before merge; the fixes from those reviews are baked in (no stuck
+  streaming cursors, robust pronunciation parsing, safe story regeneration). Clipboard writes
+  route through the main process to respect the app's locked-down renderer permissions. The
+  unit-test suite grows to 118.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-interview-assistant",
-  "version": "1.1.0",
+  "version": "1.2.0",
   "description": "BrainCue Copilot — desktop AI interview copilot (Electron + React + OpenAI). Local-first data, BYO OpenAI key.",
   "author": "tpikachu",
   "license": "MIT",