Repository navigation
fix: preserve context budgets and summary tool identity - #255
Merged
liuedcson merged 1 commit intoOct 16, 2026
Merged
Conversation
Retain exact-model context/output limits in partial overlays, preserve tool result attribution including empty results, and avoid restarting failed automatic summary prefixes when only the retained tail changes.
MelodyVAR
marked this pull request as ready for review
October 15, 2026 11:13
liuedcson
merged commit Oct 16, 2026
64154be
into
fix/harbor-then-run-bundle-20261014
1 check passed
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Partial model overlays reset known context/output limits, and summary input drops tool-call identity and error state. Preserve exact-model budgets, keep result attribution (including empty results), and prevent retained-tail growth from restarting a failed automatic summary prefix.
The automatic guard is scoped to the default CLM fallback. Changed prefixes or model/retry settings, manual compaction, overflow recovery and extension decisions retain their paths. Native history is unchanged by summary serialization.
Validation: 267 relevant tests passed, with 13 existing skips; the full native check script passed. Both rebuilt Linux ABIs passed real Astra tool-loop checks with a resolved 272000 window, 32768 output cap, max reasoning and internal review disabled. These checks do not establish a new benchmark score.
This PR targets the Harbor runtime integration branch because the released runtime also retains the pending headless interfaces from #242 and command parity from #248. The diff is limited to the context/summary repair and its tests/documentation. Release source:
9a1776d0e3dd77a094511bb1ca9fd000360a3f31.