Repository navigation
fix: drop mid-research narration segments from the rendered reply - #43
Merged
Merged
Conversation
Prompt rules reduce but cannot guarantee zero thinking-out-loud: under deep research the model still streamed "Good, that's enough confirmation. Now let me quickly check…" into the bubble (persona-loop run 11). With subagents hard-blocked hub-side, any text segment with real steps both before and after it is research narration by construction — so the reply now keeps only the opener (e.g. an empathy acknowledgement before the first step) and everything after the last step, live (applyStreamEvent), on the buffered fallback and on rehydrate (eventsToTurn). The dashboard transcript still records every block. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Robbertvermeulen
changed the base branch from
fix/chat-stream-follow-and-block-join
to
main
August 19, 2026 12:07
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Stacked on #42, one commit (9cf9881):
Prompt rules against thinking-out-loud proved probabilistic (a 20-step Peter run leaked one again), so this adds the mechanical guarantee: the rendered reply = the opener + the text after the LAST tool step; mid-research narration segments never reach the bubble. Applies to live streaming, the buffered fallback and rehydrate. The dashboard transcript keeps everything (vendor ground truth unchanged).
Verified in-browser on a real 20-step conversation.
assets/distrebuilt; JS tests green.Merge order: #42 first, then this.
🤖 Generated with Claude Code