You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Fix stale semantic web-search rows sticking at bottom after rollout commit #191
A saved debug bundle shows committed Codex rollout rows for web search already rendered in the transcript while the live proxy semantic history still renders the same web-search activity at the bottom of the feed.
html-clean.html has committed rollout rows such as 2026-05-17T12:01:20.152Z:web_search_call, 2026-05-17T12:01:37.425Z:web_search_call, and 2026-05-17T12:01:51.946Z:web_search_call.
The same HTML also has live semantic web-search rows at the bottom: 🌐 Search: Electron window.prompt... and 🌐 Search: site:electronjs.org....
render-diagnostics.json shows archived proxy turns such as resp_06c0839... and resp_0365a73... still have renderUnitCount > 0 because finalized web_search_call blocks have no toolId/callId, so committed tool ownership cannot suppress them.
Empty finalized reasoning blocks are also counted as render units even though SemanticLiveBlockRow renders null for empty reasoning.
Expected
Once Codex rollout has committed the web-search rows and final assistant text, archived proxy semantic history should not keep rendering duplicate web-search activity at the bottom.
Likely fix
Make the semantic render-unit ownership layer match actual visible rows:
empty reasoning/thinking blocks should not emit render units
finalized provider-native activity blocks without correlation ids, such as web_search_call, should yield when the same Codex turn has committed rollout ownership for those activity rows
add regression coverage based on the 2026-05-17 bundle shape
This is a follow-up to the semantic-first rendering stack and the render ownership fixes in #186.
Activity