release aidev SIG 3/5

Cline Desktop v0.0.12: per-tool transcript rows, syntax-highlighted diffs, new typography

Tool calls now each get their own transcript row with individual icons, status, and expandable detail instead of grouped summaries. File diffs render through a shared syntax-highlighted renderer. The interface switches to Inter and Geist Mono. The Claude Code provider fix from the SDK lands, and compaction respects Max Output Tokens instead of a hardcoded 1024.

PUBLISHED2026-08-13
OBSERVED2026-08-21
AGE13d
SOURCES1
  • Every tool call gets its own transcript row with icon, status, and expandable detail; commands display as $ command with captured output on expand; edits show diffs inline per hunk
  • Running tool rows highlighted in brand violet, settle to gray on completion, errors stay red
  • File diffs now render through a shared syntax-highlighted renderer following the app theme
  • Refreshed session transcript layout, message surfaces, and composer with Inter and Geist Mono typography
  • Thinking indicator stays up during quiet stretches (e.g., while tool arguments stream)
  • Message actions (copy, fork, timestamp) no longer crowd message text; expanded panels render at full opacity instead of faded
  • Welcome screen chat input is centered and top-aligned; prompt suggestions are temporarily hidden
  • First turn of a fresh session no longer wedges the composer on "Agent is working…" forever
  • Scheduled runs no longer appear in session sidebar and history list
  • Fixed reconnecting to a stale Hub daemon; prompts to update and restart, and stages the app update first
  • Hub daemon shuts down cleanly instead of exiting with error when a client is connected
  • Claude Code provider fix: sessions anchored on workspace folder, ~/.claude and project settings loaded, file edits under workspace allowed
  • Truncated tool-call JSON rejected instead of silently repaired
  • Fixed strict providers (Vercel with kimi-k3) rejecting turns with empty message content
  • Fixed mid-turn crash on streamed tool calls with non-zero indexes through LiteLLM's Anthropic passthrough
  • Compaction now respects Max Output Tokens setting instead of hardcoded 1024-token cap
  • Vertex AI: added Fable 5 and custom model IDs; global-region picker no longer hides models from live catalog
featurefix

COMMUNITY

No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.