fix(compaction): token-budget primary tail protection#6240
Closed
BongSuCHOI wants to merge 2 commits into
Closed
Conversation
Tail protection was effectively message-count based despite having a token budget, because protect_last_n=20 acted as a hard floor. A single 50K-token tool output would cause all 20 recent messages to be preserved regardless of budget, leaving little room for summarization. Changes: - _find_tail_cut_by_tokens: min_tail reduced from protect_last_n (20) to 3; token budget is now the primary criterion - Soft ceiling at 1.5x budget to avoid cutting mid-oversized-message - _prune_old_tool_results: accepts optional protect_tail_tokens so pruning also respects the token budget instead of a fixed count - compress() minimum message check relaxed from protect_first_n + protect_last_n + 1 to protect_first_n + 3 + 1 - Tool group alignment (no splitting tool_call/result) preserved
BongSuCHOI
added a commit
to BongSuCHOI/hermes-agent
that referenced
this pull request
Apr 8, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
Contributor
Author
Test Fix: context_compressor min_tail=3 ✅ pushed3 failing tests in
Root cause: PR changed min_tail from protect_last_n to min(3, ...), raising min_for_compress to 6. Tests with exactly 6 msgs hit the early return. Remaining 10 failures are pre-existing on main (quick_commands print mock, HF model, docker env, vision tools). |
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
Contributor
Author
CI Failure AnalysisVerified that all 10 test failures are pre-existing on main and unrelated to this PR. The changes here only touch context_compressor.py and its tests — none of the failing test files were modified. Test failures (all pre-existing):
build-and-push failure: pip install hits resolution-too-deep — also fails on the latest main commit. Upstream dependency resolution issue. 9329 tests passed. Safe to merge. |
teknium1
pushed a commit
that referenced
this pull request
Apr 9, 2026
PR #6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
teknium1
pushed a commit
that referenced
this pull request
Apr 9, 2026
PR #6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
Contributor
|
Merged via PR #6453 — your commits were cherry-picked onto current main with authorship preserved. Added 6 new tests covering the motivating large-tool-output scenario, min tail guarantee, 1.5x soft ceiling, and token-budget prune path. Great fix, @BongSuCHOI — this makes compaction actually effective in tool-heavy sessions! |
Contributor
Author
saxster
pushed a commit
to saxster/hermes-agent
that referenced
this pull request
Apr 9, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
Tommyeds
pushed a commit
to Tommyeds/hermes-agent
that referenced
this pull request
Apr 12, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
angelburgosrosado
pushed a commit
to angelburgosrosado/hermes-agent
that referenced
this pull request
Apr 27, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
angelburgosrosado
pushed a commit
to angelburgosrosado/hermes-agent
that referenced
this pull request
Apr 28, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
02356abc
pushed a commit
to 02356abc/hermes-agent
that referenced
this pull request
May 14, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
olympus-terminal
pushed a commit
to olympus-terminal/hermes-agent
that referenced
this pull request
May 16, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
gweeteve
pushed a commit
to gweeteve/hermes-agent
that referenced
this pull request
Jun 2, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
Egavasyug
pushed a commit
to Egavasyug/hermes-agent
that referenced
this pull request
Jun 10, 2026
PR NousResearch#6240 changed tail protection from protect_last_n to min(3, ...) which increased the minimum compressible message count and shifted tail boundaries. Three tests broke: - test_summary_role_avoids_consecutive_user_messages: 6→8 msgs - test_double_collision_user_head_assistant_tail: 7→8 msgs - test_no_collision_scenarios_still_work: 6→8 msgs All tests now exceed the new min_for_compress threshold (6) and maintain proper role alternation in both head and tail sections.
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.
Problem
Tail protection in context compaction is effectively message-count based despite having a token budget.
protect_last_n=20acts as a hard floor, so a single 50K-token tool output (file read, large API response) causes all 20 recent messages to be preserved — even if they total 200K+ tokens. This leaves almost nothing to summarize, making compaction nearly useless in long-tool-output sessions.Current behavior (200K context, 50K threshold):
If the last 5 messages total 120K tokens, all 20 are still protected → only head + tiny middle gets summarized.
Solution
Make token budget the primary criterion for tail protection, with a small message-count floor for safety:
min_tailreduced fromprotect_last_n(20) → 3 messages (hard minimum)_prune_old_tool_resultsalso respects token budget (newprotect_tail_tokensparam)Changes
Example (200K context, 50K threshold, 20K tail budget)
Before: 20 messages protected (could be 200K tokens) → almost nothing to summarize
After: ~20K tokens of recent messages protected (~5 normal msgs, or 1 large tool output + 2 msgs) → much more middle content available for summarization
Backward compatibility
_prune_old_tool_resultsnew param is optional (defaults to None → old behavior)protect_last_nstill exists as a config param, just no longer the tail floor_find_tail_cut_by_tokenssignature unchangedcontext_compressor.py