Conversation
The structural delivery change still leaves a prompt exemplar telling the model to trust marker text. It also inserts a newly arrived steer before any assistant messages following the last tool result, changing conversation chronology. Describe the runtime-owned user role without teaching a delivery-shaped marker. Keep legacy marker helpers for old histories and external callers. Append new steers at the current tail, preserving tool payloads and prior assistant output, and reuse the post-tool drain at the pre-API boundary instead of duplicating it. Follow-up to NousResearch#82467; preserves the additional provenance and prompt boundary from NousResearch#82834. Related NousResearch#81828 Co-authored-by: ygd58 <buraysandro9@gmail.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Complete the runtime-owned steer boundary in NousResearch#82467. The current branch sends user-role steers but still tells the model to trust marker text, and can place a newly arrived steer before an already delivered assistant message.
Replace that prompt exemplar with static role-based guidance, append new steers at the conversation tail, and reuse the existing post-tool drain before API calls. Preserve tool payloads, no-tool restashing, old marker helpers, and historical replay compatibility.
This preserves the additional behavior from NousResearch#82834 without maintaining a competing structural-delivery PR. It goes beyond the marker-layout-only alternative in NousResearch#82208 by removing marker-based authority from the prompt. The static prompt changes once; no per-session cache salt is introduced.
Validation: both prompt/chronology regressions fail on the exact parent
9a0d12b8c71band pass with this patch. The contributor-branch steer, segmentation, and rendered-prompt suites pass 76 tests. A separate current-main replay passes 150 tests, with one native-Windows skip, including actual pre-API delivery, legacy compression anchors, subagent steering, and bounded tool-result handling. Main moved the pre-API code intoagent/turn_iteration_prep.py; that replay preserves the new module boundary rather than restoring the old conversation loop.Not checked: live provider behavior, native Windows, full repository suite, and CodeRabbit (self-authored follow-up). This PR targets the contributor branch; the upstream PR still needs its current-main rebase.
Related NousResearch#81828
Maintenance: GPT-6 in Codex desktop (parent reasoning level unavailable). The account owner loosely reviews these actions and receives the usual GitHub notifications.