PRism Desktop v0.5.1

This release lands the long-running AI feature branch on main and turns it into something you actually control: an AI Settings tab where you can cap summary length, flip individual AI features on or off, and see a full usage-and-spend breakdown, plus a first-run onboarding dialog that walks you through picking an AI mode. Visually, AI surfaces get a new animated Prism glyph replacing the old sparkle, draft PRs are now clearly marked, and the diff view gets collapsible comment threads, fixed gutter numbers, and a wider, more readable layout. Under the hood, a token-accounting bug that badly undercounted AI usage is fixed, alongside several backend robustness hardening passes. Here's the detail. 🔷

🚀 New features

• Draft PR badge — draft pull requests are now clearly marked with an info-blue indicator on both the inbox row and the PR-detail header, sourced directly from GitHub's draft flag. (#535)
• AI usage & spend tracker — a new AI Usage view under Settings shows token totals, estimated cost, cache hit-rate, and a by-feature and by-PR breakdown across 24h/7d/30d/all windows. (#547)
• Per-feature AI toggles — independently turn Summary, File focus, Hunk annotations, and Inbox enrichment on or off from a new collapsible accordion in AI Settings. (#545)
• First-run AI onboarding — a fresh profile now greets you with a dialog to pick your AI mode (Off/Preview/Live), with Live requiring you to review and accept the data-sharing disclosure first. (#549)
• Summary length cap — a new stepper in AI Settings lets you set how long AI-generated PR summaries should be (500–5000 characters), and re-opening a summary flags it as out of date if the cap has since changed. (#533)
• Collapsible comment threads — inline review comment threads in the diff view collapse to a single line by default when resolved, expandable with one click to see the full conversation. (#574)

✨ Improvements

• Authored-by-me leads the inbox — your own open PRs now surface first by default, and a PR you authored no longer also shows up duplicated under Review requested, Needs re-review, or Mentioned. (#534)
• AI thinking indicator everywhere — the working/idle AI cue now appears consistently across summaries, file tree, Hotspots, inbox category chips, and draft suggestions, so you can always tell when AI is still processing versus done. (#551)
• New Prism AI glyph — the AI sparkle icon and its spin/pulse working animation are replaced with a custom 3D Prism-pyramid mark that tracks your chosen accent color, refined further with a follow-up weight and polish pass. (#556)
• Polished consent dialog — the "Enable Live AI" data-sharing consent modal gets a proper visual treatment with a clearer warning callout, styled buttons, and an in-progress spinner while your choice is submitted. (#539)
• Wider PR-detail columns — the Overview and Hotspots tabs now scale with your viewport instead of staying pinned to a narrow fixed width, giving noticeably less wasted margin on larger screens. (#565)
• Consistent AI summary header — the summary card header now lays out identically in Preview and Live mode, with the Sample badge properly centered. (#562)
• Diff comments match Overview styling — inline review comments in the Files/diff view now render in the same sans-serif font and centered, readable-width column as comments on the Overview tab, instead of looking cramped and monospaced. (#568)
• Cleaner AI preview chrome — the "SAMPLE" region badges above the inbox, Activity rail, and file tree are removed as visual clutter, while the inline sample markers next to actual AI output stay put. (#540)

🐛 Fixes

• Added files in the diff view now show their line numbers at rest instead of only appearing on hover. (#559)
• Switching AI mode from Live to Preview no longer leaves every eligible inbox row stuck showing the "AI working" glyph forever. (#555)
• AI token usage and cost tracking now account for Claude's cache-creation tokens, correcting a bug that had been under-reporting input token counts by orders of magnitude. (#529)

🔧 Under the hood

• Merges the long-lived experimental AI branch back into main, consolidating months of AI feature work with recent main-line changes onto a single branch going forward. (#552)
• Backend hardening: a faulty event subscriber, config-change handler, or reconcile pass can no longer take down unrelated processing, and lockfile/config internals get additional safety nets and typed error handling. (#567)

Full Changelog: v0.4.0...v0.5.1