Release 1.0.73 — composer console redesign + Git flow + dogfood set

CHANGELOG.md

@@ -4,6 +4,38 @@ Notable changes to AGBench, the local-first macOS desktop workbench for running
 and reviewing AI coding agents. Entries are user-facing highlights; execution,
 history, and workspace state stay on your machine throughout.
 
+## 1.0.73 — 2026-06-04
+
+### Added
+- **Commit & open PRs from the composer** — the Review changes menu now drives a
+  real Git flow: see your branch and changed files, write a message and Stage all
+  & Commit, then Create PR once the branch is pushed and ready (gated on a live
+  readiness check).
+- **Clearer Ensemble cost & escalation** — each round shows real vs. estimated
+  spend (a latency line + an "API-equivalent" badge on estimates), and the
+  orchestrator's escalation signals surface inline, so a multi-seat panel's value
+  is visible rather than guessed.
+- **Optional "why?" on approvals** — attach a short intent note when you allow or
+  deny an agent action; it's recorded in the approval ledger.
+
+### Changed
+- **Refined native composer** — the AGBench shell is now a cohesive console: the
+  input sits in a framed module (solid black/white outer frame, theme-tone inner
+  panel + provider rim, full-bleed and squared), the Ensemble / Create-PR / Steer
+  rows match the same solid frame, and the permission picker sits up front beside
+  the + button. Onboarding and Settings → Appearance previews reflect the new look.
+- **Deleting a chat tidies up after itself** — removing a chat now also clears
+  that chat's own run-forensic artifacts (and only those).
+
+### Fixed
+- **Kimi tool calls** — repeated calls coalesce into a single inline card that
+  updates in place (instead of stacking) and now show the target filename,
+  matching the other providers.
+- **Bug Report Refinement** — tidied how the in-app reporter shows your workspace
+  (now a friendly `~/…` label).
+- Onboarding, empty states, the welcome dashboard, provider accent colours, and
+  the Diff Studio / File Editor light themes all got polish + readability fixes.
+
 ## 1.0.72 — 2026-06-04
 
 ### Security

package.json

@@ -1,6 +1,6 @@
 {
   "name": "agbench",
-  "version": "1.0.72",
+  "version": "1.0.73",
   "description": "AGBench desktop workspace",
   "main": "./out/main/index.js",
   "author": "Chris Izatt",

src/main/AppStoreDeleteChat.test.ts

@@ -0,0 +1,104 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+import fs from 'fs'
+import { join } from 'path'
+import { AppStore } from './store'
+import type { ChatRecord, ChatRun } from './store/types'
+
+const userDataPath = vi.hoisted(() => `/tmp/agentbench-delete-chat-test-${process.pid}`)
+
+vi.mock('electron', () => ({
+  app: {
+    getPath: () => userDataPath
+  }
+}))
+
+const runEventPath = (runId: string): string => join(userDataPath, 'run-events', `${runId}.jsonl`)
+const artifactDir = (runId: string): string => join(userDataPath, 'run-artifacts', runId)
+
+function makeRun(runId: string): ChatRun {
+  return { runId, startedAt: '2026-05-08T00:00:00.000Z' }
+}
+
+function seedRunFiles(runId: string): void {
+  fs.mkdirSync(join(userDataPath, 'run-events'), { recursive: true })
+  fs.writeFileSync(runEventPath(runId), `{"runId":"${runId}"}\n`, 'utf8')
+  fs.mkdirSync(artifactDir(runId), { recursive: true })
+  fs.writeFileSync(join(artifactDir(runId), 'stdout.log'), 'stream\n', 'utf8')
+}
+
+function saveChatWithRuns(appChatId: string, runs: ChatRun[]): ChatRecord {
+  const chat: ChatRecord = {
+    appChatId,
+    scope: 'workspace',
+    chatKind: 'single',
+    provider: 'gemini',
+    title: appChatId,
+    workspaceId: 'workspace-1',
+    workspacePath: '/repo',
+    createdAt: 1,
+    updatedAt: 1,
+    archived: false,
+    messages: [],
+    runs
+  }
+  AppStore.saveChat(chat)
+  return chat
+}
+
+describe('AppStore.deleteChat run cleanup', () => {
+  beforeEach(() => {
+    fs.rmSync(userDataPath, { recursive: true, force: true })
+    fs.mkdirSync(join(userDataPath, 'chats'), { recursive: true })
+  })
+
+  it('removes the deleted chat run-event files and artifacts', () => {
+    saveChatWithRuns('chat-a', [makeRun('run-1'), makeRun('run-2')])
+    seedRunFiles('run-1')
+    seedRunFiles('run-2')
+
+    expect(fs.existsSync(runEventPath('run-1'))).toBe(true)
+    expect(fs.existsSync(artifactDir('run-1'))).toBe(true)
+
+    AppStore.deleteChat('chat-a')
+
+    // Chat JSON gone (behaviour preserved).
+    expect(fs.existsSync(join(userDataPath, 'chats', 'chat-a.json'))).toBe(false)
+    // Both runs' forensic files removed.
+    expect(fs.existsSync(runEventPath('run-1'))).toBe(false)
+    expect(fs.existsSync(artifactDir('run-1'))).toBe(false)
+    expect(fs.existsSync(runEventPath('run-2'))).toBe(false)
+    expect(fs.existsSync(artifactDir('run-2'))).toBe(false)
+  })
+
+  it('leaves a sibling chat with a prefix-similar run id untouched', () => {
+    // chat-a owns `run-1`; sibling chat-b owns `run-1-extra` whose id has
+    // `run-1` as a string prefix. A prefix/readdir-based delete would wrongly
+    // catch the sibling's files; an exact-name delete must not.
+    saveChatWithRuns('chat-a', [makeRun('run-1')])
+    saveChatWithRuns('chat-b', [makeRun('run-1-extra')])
+    seedRunFiles('run-1')
+    seedRunFiles('run-1-extra')
+
+    AppStore.deleteChat('chat-a')
+
+    // Deleted chat's run is gone...
+    expect(fs.existsSync(runEventPath('run-1'))).toBe(false)
+    expect(fs.existsSync(artifactDir('run-1'))).toBe(false)
+    // ...but the sibling's prefix-similar run is fully intact.
+    expect(fs.existsSync(runEventPath('run-1-extra'))).toBe(true)
+    expect(fs.existsSync(artifactDir('run-1-extra'))).toBe(true)
+    expect(fs.existsSync(join(userDataPath, 'chats', 'chat-b.json'))).toBe(true)
+  })
+
+  it('succeeds when a run-event file is already missing', () => {
+    // run-1 has files, run-2 was never persisted (missing on disk).
+    saveChatWithRuns('chat-a', [makeRun('run-1'), makeRun('run-2')])
+    seedRunFiles('run-1')
+    expect(fs.existsSync(runEventPath('run-2'))).toBe(false)
+
+    expect(() => AppStore.deleteChat('chat-a')).not.toThrow()
+
+    expect(fs.existsSync(runEventPath('run-1'))).toBe(false)
+    expect(fs.existsSync(join(userDataPath, 'chats', 'chat-a.json'))).toBe(false)
+  })
+})

src/main/AppStoreSettings.test.ts

@@ -0,0 +1,44 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest'
+import fs from 'fs'
+import { AppStore } from './store'
+
+const userDataPath = vi.hoisted(() => `/tmp/agbench-settings-test-${process.pid}`)
+
+vi.mock('electron', () => ({
+  app: {
+    getPath: () => userDataPath
+  }
+}))
+
+describe('AppStore settings defaults', () => {
+  beforeEach(() => {
+    fs.rmSync(userDataPath, { recursive: true, force: true })
+    fs.mkdirSync(userDataPath, { recursive: true })
+  })
+
+  it('defaults packaged update checks to the stable channel', () => {
+    expect(AppStore.getSettings().updateChannel).toBe('stable')
+  })
+
+  it('normalizes persisted changelog metadata on load', () => {
+    AppStore.updateSettings({
+      lastSeenChangelogVersion: ' 1.0.72 ',
+      pendingUpdateChangelog: {
+        version: ' 1.0.73 ',
+        releaseName: ' AGBench 1.0.73 ',
+        releaseDate: ' 2026-06-04T12:00:00.000Z ',
+        releaseNotes: [{ version: ' 1.0.73 ', note: 'Updater UI.' }, { version: '', note: '' }]
+      }
+    })
+
+    expect(AppStore.getSettings()).toMatchObject({
+      lastSeenChangelogVersion: '1.0.72',
+      pendingUpdateChangelog: {
+        version: '1.0.73',
+        releaseName: 'AGBench 1.0.73',
+        releaseDate: '2026-06-04T12:00:00.000Z',
+        releaseNotes: [{ version: '1.0.73', note: 'Updater UI.' }]
+      }
+    })
+  })
+})