chore: clean rollout extraction in memories #11471
Conversation
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 42d3061168
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| if let RolloutItem::ResponseItem(item) = item | ||
| && should_persist_response_item(item) |
There was a problem hiding this comment.
The new filter only retains RolloutItem::ResponseItem, which drops RolloutItem::Compacted entries from the stage-1 memory prompt. For long threads this regresses extraction quality: build_stage_one_input_message truncates rollout payloads, and compacted entries are the condensed representation of earlier context that survives near later turns, so excluding them can make stage-1 memory output miss pre-compaction facts. The previous implementation converted compacted items into assistant ResponseItem::Message values before serialization, so this is a functional behavior loss rather than a refactor-only change.
Useful? React with 👍 / 👎.
There was a problem hiding this comment.
That's fine, they don't have interesting content anyway
1 participant
No description provided.