AgentLauncher v0.2.0

0.2.0 - 2026-08-17

The first formally versioned release of AgentLauncher — a Qt6/QML + C++
desktop app that launches the web UI of AI coding agents from a single
card grid. Everything is config-driven: agent definitions live in
agents.json, never in C++.

Added

• Card grid home screen showing all configured agents, each with a
  Start button (launches the agent's web server) and a Configure button
  (opens the agent's settings page).
• HTTP health-check state detection: running cards are highlighted
  with the agent's own color and a colored border; any HTTP response means
  running, connection refused/timeout means stopped.
• Stop button (×) on running cards for agents started this session —
  launch() records the PID from startDetached, and stop() runs
  taskkill /F /T /PID (Windows) so the whole cmd → .cmd → node tree
  dies.
• Transient launching state with a 30s safety timeout: the card
  shows a spinner until the health check confirms the server is up.
• Launch error feedback: launchFailed signal triggers an at-place
  red flash on the card plus a scrollable, monospace central error popup.
• Configurable agent icons and colors via agents.json: icons accept
  qrc:/ resources, local file paths (with %VAR% and ~ expansion),
  http(s):// URLs, or empty (falls back to built-in default); colors are
  auto-assigned from a built-in Catppuccin Mocha palette when empty, and an
  optional cardColor sets the non-running background.
• Four neutral built-in icons: default, terminal, cube, bot.
• Install / update / version support: each agent can declare
  installCommand, updateCommand, and versionCommand; card version
  labels show the installed agent version.
• Streaming command output to the card: while an agent is installing,
  updating, or running its one-time setup, a scrollable monospace console
  below the status line shows live stdout/stderr. It hides on success,
  stays visible for 5s on failure, is dismissible, and can be re-opened
  via a right-click "Show output" action.
• One-time setupCommand: runs before the first launch of an agent
  (e.g. generating a bearer token for qwen serve). On exit code 0 the
  result is persisted to agent_state.json and never re-run unless the
  user picks "Re-initialize" from the card's context menu.
• Bearer token authentication via the tokenFile field (Qwen Code):
  the token is set as QWEN_SERVER_TOKEN on launch and appended as
  #token=<value> to the web URL.
• Force Stop context-menu action: kills the process listening on an
  agent's web port, even when this launcher didn't start it.
• Runtime version badges in the top-right corner: detected Python and
  Node.js versions are shown as green badges; missing runtimes show a red
  × with a tooltip explaining affected agents may not work.
• Window title config: an optional title field in the agents.json
  root overrides the application window title; empty/absent falls back to
  AgentLauncher.
• Rotating file logger: all Qt log output is written to
  ~/.AgentLauncher/log/agentlauncher.log, rotating at 10 MB keeping
  2 files (current + 1 backup). Existing qWarning() calls are captured
  automatically.
• Window close confirmation: when agents have been launched, closing
  the window prompts whether to terminate all background processes started
  this session.
• Default agents: Kimi Code, OpenCode, Qwen Code, OpenClaw, and
  DeepSeek Harness ship in the bundled default_agents.json.
• Windows packaging script (scripts/package.sh): one-command
  Release build + windeployqt. Double-clickable from Explorer — it
  auto-loads vcvars64.bat when the MSVC environment is missing, always
  cds to the project root, and uses an explicit windeployqt.exe path.
• Application icon (app.rc, app-icon.png).
• Internationalization (i18n): source strings are English; a
  QTranslator auto-loads Chinese (agentlauncher_zh_CN.qm) based on
  system locale. .ts sources are synced via lupdate and compiled to
  .qm embedded under :/i18n/.
• Documentation site (MkDocs + Material, English + 中文) with a
  configuration guide, plus main-window screenshots in the README/docs.

Changed

• Launch reliability: bare commands are resolved through
  QStandardPaths::findExecutable (which applies PATHEXT), and .cmd/
  .bat shims run via cmd /c so npm-style agents (e.g. qwen.cmd)
  launch correctly — CreateProcess alone can't find them.
• Qwen Code command simplified to qwen serve; the bearer token is
  handled via tokenFile + setupCommand instead of being inline.
• Context menu stability: Force Stop and Re-initialize items now use
  enabled (greyed out) instead of visible, so the menu no longer
  grows/shrinks as state changes — consistent with Update/Install and
  Show output.
• Card layout: the button row is anchored to the bottom of the card
  rectangle, eliminating the large empty gap left by top-stacked Column
  content.
• Version-check UX: a 500ms minimum spinner duration so the indicator
  is always visible; checkingVersion is initialized before QML renders
  so cards show the spinner from the first frame; stderr is read as a
  fallback for version extraction, and a non-zero exit with a parseable
  version string is treated as installed.

Fixed

• ConfigPage property collision: the data property was renamed to
  agentData to stop shadowing QQuickItem.data, which had left fields
  empty by resolving child bindings to the wrong object.
• Card Flow overflow: the 4th card (OpenClaw) overflowed off the
  right edge because ColumnLayout width was bound to an undefined
  parent.availableWidth (ScrollView's internal Flickable has none);
  rebound to scrollView.availableWidth so cards wrap and reflow on
  resize.
• Install/update state stuck: removed the trailing & pause from
  install/update commands (it waited for a keypress so
  QProcess::finished never fired, leaving the card stuck on
  "Installing…"); added running-protection (reject while the agent is
  running) and correct installFinished signal emission for both success
  and failure.
• Stop button state: the stopping state now resets when stop()
  fails, so the button no longer stays stuck.