Persist direct-provider tool calls, results, display details, and turn identities so saved conversations rebuild individual panes. Keep restored runs dormant until explicitly resumed and bind queued shell commands to their actual execution blocks.
Persist exact provider history and archived tool results independently of
terminal scrollback, and retain them when active run checkpoints are cleared.
Keep conversation and session saving independent of automatic reopening.
Deduplicate only committed session snapshots, retry transient SQLite write
contention, and enqueue conversation updates before writer shutdown. Recover
healthy startup data independently and protect unread records by disabling
writes with a persistent warning after an incomplete restore. Retain local
transcript pane conversation links across restarts.
Add regression coverage for SQLite failures, context round trips, legacy
fallbacks, and transcript snapshots.
Bring over selected Warp UI improvements: selectable Thought text, message
timestamps, animated theme backgrounds, editable New Window shortcuts,
safer tab renaming and restoration, and text/font/discard crash fixes.
Match upstream orchestration pill geometry and tooltips, and replace expanded
subagent previews with compact links that retain live status and conversation
navigation for launches and follow-up messages.
Include the pending Cargo.lock dependency refresh.
Adapted upstream PRs: #15948, #15884, #15691, #14618, #15772, #15705,
#15605, #14895, #15996, #15771, #14893, and #14603.
Allow new provider runs to continue without an arbitrary lifetime turn budget while preserving explicit limits, checkpoint compatibility, and bounded retries. Keep pending tool proposals visibly queued until execution status arrives, and show cancellation only when supported by actual state.
Validation: formatting, inline test-layout checks, both presubmit Clippy commands, 325 targeted regressions, and 54 core tests after final cleanup passed.
Preserve full terminal-grid width with horizontal scrolling in tool panes. Keep hidden command control changes from unpinning the viewport, and re-pin scrolling at the live edge in normal and inverted layouts.
Validation: formatting, test-layout checks, both presubmit Clippy commands, and 141 relevant tests passed. Ten other terminal-view tests failed identically on the untouched baseline (keyboard, cloud-agent, input status, and footer behavior).
Wait for shell bootstrap before dispatching child prompts, serialize provider preprocessing, and make permission callbacks idempotent. Restrict task lists to concrete multistep plans and stop inferring completion from tool activity.
Update regression fixtures and resolve existing lint and test-layout failures. Verified formatting, both presubmit Clippy commands, and 354 targeted nextest tests.
rkw6086
Ryan Ward