Repository navigation
v2.57.7 — the in-progress mark works again after /wrapup and /handoff
Fixes the known issue listed in 2.57.5 and 2.57.6: the in-progress mark works again after /wrapup and /handoff. 2.57.5 stopped the first-edit hook from rewriting a SESSION.md whose ## Next session prompt holds a live handoff. Its emptiness test counted fence lines as content, so the empty fence pair that /wrapup and /handoff both leave in that slot read as a live prompt. After either one, the next session's first edit was never marked in progress (measured on an installed 2.57.5: an empty-fence file stayed wrapup, the same file without a fence was marked). Nothing was overwritten or lost.
kt_ss_read_live_promptdrops fence-only lines before testing for emptiness. An empty fence, a fenced(session in progress)and a fence with an info string all count as empty; a fence around real text is still live.- Tests: O6 (an empty fenced slot is marked in progress) and V5 (
consume_activeleaves an empty fence byte-identical), both seen red on the 2.57.5 library. Every earlier empty-slot fixture had no fence, which is why the suite was green. Also checked on copies of two real affectedSESSION.mdfiles (now marked) and a real waiting handoff (left byte-identical). - Ports: the same function in Cursor and Codex and Antigravity. Codex never calls the hook, so the change only affects
consume_activethere.
Versions: Claude Code 2.57.7, Antigravity 2.57.7, Codex 2.57.7-codex.0 and Cursor 2.57.7-cursor.0. Cowork is unchanged at 1.8.1.