What's Changed
AI-Powered Review Pane
- New: Group review — Navigate to a semantic group in the sidebar to see an AI-generated review with four sections: WHY (intent), WHAT (impact table), HOW (flow diagram), and VERDICT (high-severity issues with bug numbers like RV-1, RV-2)
- Parallel pre-computation — Reviews for all groups are spawned automatically when grouping completes, so results are ready by the time you navigate
- Two-tier caching — Reviews are cached in-memory (LRU, max 20) and on disk (
.git/semantic-diff-cache/reviews/), surviving across sessions - Review SKILL discovery — Detects custom review SKILLs from
.claude/skills/(codebase) or~/.claude/skills/(global) and injects them into the Verdict prompt - Review source banner — Shows which SKILL powered the review, or a tip to create one
- Keyboard controls —
Rto force-refresh,Escto dismiss,j/k/g/G/Ctrl-d/Ctrl-uto scroll
Markdown & Diagram Improvements
- Table row separators — Dashed lines (┄) between data rows for better readability
- Mermaid image sizing — Small diagrams no longer stretch to fill wide terminals; images are capped at their natural width
Bug Fixes
- Fix deadlock risk — stdout and stderr are now read concurrently when invoking Claude CLI
- Fix stale review cache — Group content hash now includes hunk indices, not just file paths
- Fix error caching — Reviews with failed sections are not persisted to disk (retryable on next run)
Install / Upgrade
cargo install semantic-diff
# or
brew install alankyshum/tap/semantic-diffFull Changelog: v0.8.0...v0.9.0