Added
- Codex and Claude skill activity stays visible above folded chat work blocks: the skill name,
invocation/read status, and expandable evidence or document path. Recorded activity is distinct
from completing the skill's workflow; English and Korean labels work on desktop and mobile. - Native Codex image attachments, including image-only prompts, appear in the chat through
bounded, pane-scoped image reads.
Fixed
- Native context clears discard old turns, pending calls, loaded pages and stale cursors.
Late page and tool-output responses cannot restore cleared history or populate another pane. - omp, omo and gjc transcripts honor hidden messages and normalize string messages, tool field
aliases and embedded results consistently across rendering, paging and full-output reads. - omo transcript selection uses process/session evidence and rejects ambiguous same-directory
candidates. Claude paste wrappers unwrap only when their identifiers match.
Changed
- Growing Codex tasks parse incrementally while preserving transcript rewrite invalidation,
reducing repeated parsing of long tool-heavy turns.