Repo-audit cleanup + remove legacy path/training + fix eject

.env.example

@@ -51,3 +51,7 @@ ELEVENLABS_API_KEY=
 # AI_VIDEO_MODEL=
 # Where generated video clips are written on disk (default: .data/video/):
 # A2UI_VIDEO_DIR=
+# Where generated music beds are written on disk (default: .data/music/):
+# A2UI_MUSIC_DIR=
+# Where generated TTS recordings are written on disk (default: .data/tts/):
+# A2UI_RECORDING_DIR=

.github/workflows/ci.yml

@@ -6,6 +6,12 @@ on:
   pull_request:
     branches: [main]
 
+# Cancel a PR's superseded in-flight runs when a new commit is pushed; main runs
+# are keyed by ref and never auto-cancelled.
+concurrency:
+  group: ci-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
 jobs:
   check:
     runs-on: ubuntu-latest
@@ -28,9 +34,9 @@ jobs:
           cache: "pnpm"
 
       - name: Install dependencies
-        run: pnpm install
+        run: pnpm install --frozen-lockfile
 
-      - name: Run checks (prettier, eslint, vitest, build)
+      - name: Run checks (prettier, eslint, stylelint, typecheck, vitest, build)
         run: pnpm check
 
   # End-to-end Playwright specs. Non-blocking (continue-on-error) and the visual
@@ -42,9 +48,9 @@ jobs:
     runs-on: ubuntu-latest
     needs: check
     continue-on-error: true
-    # Hard backstop: e2e serves a production build (see playwright.config.ts) and
-    # runs ~100 specs; this caps a hung run rather than letting it idle for the
-    # default 6h. Comfortably above the expected build + suite time.
+    # Hard backstop: e2e runs the suite against a `next dev` server (see
+    # playwright.config.ts), which cold-compiles each route on first hit; this
+    # caps a hung run rather than letting it idle for the default 6h.
     timeout-minutes: 20
 
     strategy:
@@ -64,7 +70,7 @@ jobs:
           cache: "pnpm"
 
       - name: Install dependencies
-        run: pnpm install
+        run: pnpm install --frozen-lockfile
 
       - name: Install Playwright browsers
         run: pnpm exec playwright install --with-deps chromium

.gitignore

@@ -6,6 +6,10 @@
 /.pnp
 .pnp.*
 
+# pnpm is the only package manager (see packageManager field); the npm lockfile
+# must never be committed — a stale one once drifted from pnpm-lock.yaml.
+package-lock.json
+
 # pnpm content-addressable store (created locally when store-dir resolves to
 # the repo; must never be committed — prettier --check walks it and CI fails)
 .pnpm-store/

.husky/pre-commit

@@ -1 +1 @@
-pnpm check
+pnpm exec lint-staged

.nvmrc

@@ -0,0 +1 @@
+24

AGENTS.md

@@ -251,4 +251,4 @@ pnpm e2e tests/e2e/a2ui/
 
 ### Legacy Protocol
 
-The original tool-result protocol (`src/lib/protocol/schema.ts`, 23 component types) is still functional but superseded. New features should use A2UI v0.9.
+The original tool-result protocol (`src/lib/protocol/schema.ts`, 24 component types) is still functional but superseded. New features should use A2UI v0.9.

README.md

@@ -5,7 +5,7 @@
 An [A2UI Protocol](https://a2ui.org/) showcase built as a noir-themed AI interface: generated UI arrives with a matching palette, typography, voice, and soundtrack. The chat → generate → eject-to-code loop draws on tools like [Lovable](https://lovable.dev) and v0, but the rest of the app is its own thing.
 
 [![CI](https://img.shields.io/badge/CI-passing-brightgreen)]()
-[![Tests](https://img.shields.io/badge/tests-1193%20passing-brightgreen)]()
+[![Tests](https://img.shields.io/badge/tests-1200%2B%20passing-brightgreen)]()
 
 ![Detective workspace with evidence board and chat](docs/screenshots/noirable_screen.jpg)
 
@@ -20,7 +20,7 @@ synthNoirUI is a full-stack reference implementation of the **A2UI (Agent to UI)
 The project implements two protocol versions:
 
 - **A2UI v0.9** -- 18 components in the standard catalog, streamed via SSE as JSONL messages (`createSurface`, `updateComponents`)
-- **Legacy protocol** -- 23 component types delivered through Vercel AI SDK tool calls, validated with Zod schemas
+- **Legacy protocol** -- 24 component types delivered through Vercel AI SDK tool calls, validated with Zod schemas
 
 Both protocols produce the same visual output. The v0.9 path is the active development target.
 

docs/DEVELOPMENT.md

@@ -188,7 +188,7 @@ The Vitest thresholds in `vitest.config.ts` are the gate:
 - Functions: **65%**
 - Statements: **70%**
 
-Run `pnpm test:coverage` to check. The suite currently has 875 tests across 91 test files.
+Run `pnpm test:coverage` to check. The suite currently has 1,200+ tests across 100+ test files.
 
 ---