Release v0.7.0 · lynnswap/CodexReviewKit

CodexReviewKit v0.7.0

ReviewMonitor now retains completed, failed, and cancelled review history across app relaunches, including targets, models, outcomes, final review text, findings, and sidebar ordering. The history database keeps up to 50 reviews per workspace and 500 overall, and does not persist raw transcripts, reasoning, command output, or credentials. History begins with reviews run on v0.7.0.
Added secure API-key sign-in from the signed-out screen and the Add Account menu.
Added authoritative typed terminal outcomes to the Swift API and MCP lifecycle responses, distinguishing completion, failure, requested interruption, server interruption, transport failure, and previous-process exit.

Long review histories now support Show more/Show less, multi-select deletion, and durable reordering across primary checkouts and linked worktrees.
ReviewMonitor now preserves every reasoning effort advertised by Codex—including Max, Ultra, and future values—marks high-usage options, and uses compact model names in the sidebar.
Internal cancellation and cleanup diagnostics stay out of normal ReviewMonitor and MCP logs while remaining available through review_read with logFilter: "all".

Updated current Codex app-server event and interrupt handling so reviews keep the canonical final result, cancellation targets the active review turn, and incomplete or disconnected streams are not reported as successful.
Preserved in-flight settings changes across runtime replacement and prevented stale authentication callbacks or cancellation from overwriting newer account state.
Improved Codex executable discovery for Finder launches and common local installations when the GUI environment has a minimal PATH.
Kept bottom-pinned logs attached during resize and reflow, made log-block spacing consistent, and stopped showing misleading 0s durations for completed subsecond commands.

Full Changelog: v0.6.2...v0.7.0