real dev verification: PR 23940 main: 15fe2832727893cbdd2214383f9f971474849695 head: 5b710f34f3e60b58a173b52d780004f04c96987d Chrome: /usr/bin/google-chrome; Playwright: repo dependency 1.60.0 main server/web: 16733/8693; stopped before PR server/web: 13773/5733 Both projects added through UI, pointing at /home/user/pr23940-scenario (real git repo). main opening-last result:
`test` after. reply with exactly BACKTICK_OK, no tools.
main persisted run: {"provider":"codex","model":"gpt-6.1-sol","status":"completed"} [{"role":"user","text":"`test` after. reply with exactly BACKTICK_OK, no tools.","source":"web"},{"role":"assistant","text":"BACKTICK_OK","source":"provider"}] PR persisted run: {"provider":"codex","model":"gpt-6.1-sol","status":"completed"} [{"role":"user","text":"`test` after. reply with exactly BACKTICK_OK, no tools.","source":"web"},{"role":"assistant","text":"BACKTICK_OK","source":"provider"}] 8.3s verified model picker: GPT-6.1-Sol (Codex) 8.7s recording scenario starts 13.9s opening backtick typed last: code mark applied 15.1s undo: original closing-only text restored 16.1s redo: code mark restored 17.6s caret stayed at code start: new text inserted inside code 18.4s plain continuation stayed outside the code mark 21.6s neighbor: ordinary closing-last code still works 23.6s neighbor: doubled closing backticks remain literal 27.9s final submission composer HTML:test after. reply with exactly BACKTICK_OK, no tools.