GitDesktop v0.5.0

Added

• Re-run a stopped automated review. When an automation-triggered AI review or
  security audit is cancelled or fails, it now stays in the Activity popover under a
  Stopped group with Re-run and Dismiss, so you can retry exactly that run
  (and mode) without hunting for the commit or PR again. A failed automated run also
  lands a review failed notification in the inbox — matching manual runs — when
  automation notifications are on.
• Create pull requests as drafts by default. A new Settings → General toggle,
  Create pull requests as drafts (off by default), pre-checks the Create pull request
  dialog's Create as draft box so you can open PRs as drafts without remembering to tick
  it each time — still overridable per PR. Paired with Review draft PRs when created, a
  draft's automated first review then waits until you mark it ready.
• See how long reviews take. The Activity & notifications popover now shows a live
  elapsed timer on a running AI review, notes how long a stopped (cancelled or failed)
  automated run ran before it stopped, and the Previous reviews list shows each
  finished review's total run duration. The PR review panel also ticks a live elapsed
  timer beside Cancel while a review runs, and shows the total time it took once it's done.
• Notes for reviewers. Hand review context to the AI reviewer: an agent (or any
  MCP client with write access) deposits per-branch notes via the GitDesktop MCP, and
  the Create pull request dialog shows an optional Notes for reviewers field that
  pre-fills from that deposit for the head branch. On create the notes are posted as the
  PR's first comment and fed to the automated review as first-class context, so
  deliberate, documented decisions stop getting re-flagged. A new Review draft PRs when
  created automation setting (off by default) holds a draft PR's first review until it's
  marked ready for review.
• Dedicated security-audit model. Give AI security audits their own provider and model,
  separate from the general review model — e.g. a stronger model for audits and a faster one
  for everyday reviews. Toggle Use a different model for security audits under the review
  model in Settings → AI; left off, audits keep using the review model exactly as before. The
  choice applies to both automated audits and the Security audit button on a PR (an
  in-panel model pick still overrides both for that one run).
• Tasks — save scripts and run them in-app. A new Tasks tab (and a
  command-palette Run a task…) lets you register your own scripts — a release or
  build flow, say — and run them without dropping to a terminal. Point a task at an
  existing script in the repo (it runs the live file, so edits take effect on the next
  run) or write one inline; with an AI provider connected, Generate writes an
  inline script from a plain description, and Analyze with AI reads a script and fills
  in its name, description, and the arguments it accepts. Each task carries a description
  and default arguments (quoted values kept intact), documents its arguments
  --help-style, and a confirm-gated run lets you adjust the arguments per run. Runs
  happen in an interactive terminal in the repository's folder, so scripts that prompt
  you (a version to release, a yes/no) work and keep their colour; Stop kills the run
  and its child processes, Rerun starts a fresh one. Choose the interpreter — PowerShell,
  cmd, Git Bash, bash/sh/zsh, Node, Deno, Bun, Python, or Ruby, with the editor showing which
  it detected on your machine. Task definitions live in your app data
  and are never read from repository content, running is off until you enable it, and each
  task can ask for confirmation before it runs.
• Themes. Choose how GitDesktop looks in Settings → Appearance: System
  (follow your OS), Light, Dark, or a softer Slate — a cool, lifted
  blue-gray that eases eye strain on long sessions. Cycle themes from the command
  palette too.

Changed

• CI run and job log lookups now thread ids as strings end-to-end, so they stay
  precise even above JavaScript's safe-integer limit (2^53). The MCP workflow
  tools accept a run/job id as either a number or a numeric string, so existing
  callers keep working.
• Toggle a PR between draft and ready on every provider. The pull-request footer's
  Ready for review / Convert to draft pair now works both ways on GitHub,
  GitLab, and Bitbucket — not just Bitbucket, and no longer one-directional. Both
  actions are also in the command palette, and the GitDesktop MCP set_pull_request_draft
  tool now covers all three providers too.

Fixed

• Ignore and AI-exclude toasts now report how many patterns were actually
  added, rather than the size of the selection — so a fully-covered selection
  says everything was already ignored instead of claiming entries were added.
• Publishing a draft release no longer clears its Latest status. Publishing
  a draft from the app now follows GitHub's default and becomes the Latest
  release on publish, instead of being forced non-latest. The Edit dialog explains
  that Latest applies only to published releases rather than offering a toggle
  GitHub would silently ignore on a draft. The MCP update_release tool had the
  same issue and is fixed too.
• Fixed AI features failing when their input text contained an emoji near a size
  limit. When a prompt was truncated to fit its budget, the cut could split an
  emoji (a UTF-16 surrogate pair) and leave an invalid Unicode fragment that the
  model provider rejected ("unexpected end of hex escape" / "Invalid body") — most
  visibly breaking AI PR reviews when a bot review or prior comment contained an
  emoji, but the same flaw affected PR-description and commit generation, issue
  drafting, merge-conflict resolution, and repo/discussion prompts. Prompt
  truncation now respects character boundaries everywhere, so emoji are never split.
• User-facing copy no longer hardcodes "right-click": on macOS the in-app guide
  and branch-status tooltips now say Control-click, matching how context
  menus actually open on trackpads and swapped-button mice. References to "its
  right-click menu" are now the platform-neutral "context menu" — across the
  in-app guide, README, and marketing site.
• Terminal and editor auto-detection now works on macOS and Linux, not just
  Windows. Settings → Terminal / External editor populate with the terminals and
  editors you actually have installed (Terminal, iTerm, Warp, Ghostty, VS Code,
  Cursor, Sublime Text, Zed, Xcode, JetBrains IDEs, and more), and "Open in
  terminal" / "Open in editor" now honor the one you pick instead of always
  falling back to the system default.