fix: stop extra narration after an agent already replied while its sub-agents run (#200655) * fix(agents): preserve waiting replies for already-pending children (cherry picked from commit bc0ca8a5537746d6c1f5bf4ac073b37ecf36ff8b) * fix(agents): end the turn when an already-pending yield follows a delivered reply * refactor(agents): count a waiting acknowledgment as yield continuation evidence * test(ui): prove an already-pending yield after a delivered reply ends the turn * test(ui): drive the already-pending yield e2e through message-tool-only replies * test(ui): accept message-tool-only completion input in the wait fixture * fix(agents): require the current input's own delivered reply before an already-pending yield ends it * fix(agents): end an already-pending yield turn only when a pending child will wake the session * style(agents): bind the steered-input reset and test invoker explicitly * test(ui): register the already-pending yield e2e as a real-Gateway file and allocate its test lines * test(ui): list the already-pending yield e2e with the serial real-Gateway files * test: account for the inline yield acknowledgment case after the suite fold --------- Co-authored-by: Jan Leon <jan@leoncode.de>
fix: stop extra narration after an agent already replied while its su…
Find Your Interest
- improve(ci): avoid duplicate release source guard scans (#200379)
- fix: accept first chat while the model catalog loads (#201127)
- fix: let hooks start the first agent session (#201103)
- fix: let first sessions save labels and models (#201022)
- docs: require latest bundled Codex in release preparation (#201147)
- fix: keep optional release tooling comparisons advisory (#200527)
- fix: expose saved dashboards in session inventory
- docs: clarify Codex upgrades for newer models (#201003)
- fix: enforce publication fence budgets across worker clocks (#200518)
- refactor: keep outbound message chunk limits consistent (#201132)





