Repository navigation
fix(compress): hand @session and @memory output to the model whole - #1869
Merged
Merged
Conversation
The compressors pick a strategy from the content, and a saved conversation that talks about errors, build steps or file paths reads like a log. A session opened to answer a question came back with the middle dropped behind a lines-omitted marker, so the evidence could be cut before the model saw it. Both tools now join @recall in passing through verbatim. On LongMemEval-S the cut hit 68 of 500 questions, and those scored 84% against 91% for the rest. With the change it hits 1 (a @tools listing).
Contributor
Quality GateResult: ✅ all floors passed
Config: .github/quality-gate.yml. Workflow: |
This was referenced Oct 22, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
@sessionand@memoryresults are no longer run through the content-aware compressors. They join@recall, which already passed through verbatim.Why
The compressors choose a strategy from the content. A saved conversation that mentions errors, build steps or file paths reads like a log, so the middle of a session the model opened to answer from came back as
... [152 lines omitted] .... The evidence could be cut before the model saw it, with no signal other than the marker.Measured on LongMemEval-S (500 questions, same dataset and judge as the 2026-10-20 run):
@tools describelisting, unrelated). Those 68 now score 58/68.Tests
TestLayerNeverCompressesMemoryToolOutput: 300 session-like lines compress to 3104 bytes under a log hint, and pass through byte-identical for@session,session,@memory,Memory. Fails on main (strategylog), passes here.go test ./cli/...green.