🤖 feat: force compaction when approaching context limit #778
Conversation
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
ethanndickson
force-pushed
the
force-compaction-interrupt
branch
from
41d3661 to
22443ea
Compare
Add force compaction that triggers automatically when live token usage approaches the context window limit during streaming. This prevents the AI from hitting context limit errors mid-response. Key changes: 1. Auto-compaction check extended with shouldForceCompact flag - Uses liveUsage (real-time streaming) or falls back to lastUsage - Triggers when remaining tokens <= FORCE_COMPACTION_TOKEN_BUFFER - Works even with empty usageHistory (first message streaming) 2. AIView triggers force compaction during active streams - Tracks triggered stream to prevent duplicate compactions - Sends compaction request with 'Continue with current task' message 3. Queue restore moved from stream-abort to IPC interrupt handler - Prevents continueMessage from bouncing back to input on internal aborts - User-initiated interrupts still restore queued messages to input 4. Shared constants for compaction parameters - DEFAULT_COMPACTION_WORD_TARGET = 2000 - WORDS_TO_TOKENS_RATIO = 1.3 - FORCE_COMPACTION_TOKEN_BUFFER = 5200 (2x expected output)
ethanndickson
force-pushed
the
force-compaction-interrupt
branch
from
22443ea to
ea4d8fc
Compare
Member
Author
|
@codex review |
|
Codex Review: Didn't find any major issues. Keep them coming! ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
github-merge-queue bot
pushed a commit
that referenced
this pull request
Dec 1, 2025
## Summary Automatically triggers compaction when live token usage approaches the context window limit during streaming. This prevents the AI from hitting context limit errors mid-response. ## Changes ### Force Compaction Logic - Extended `checkAutoCompaction()` to return `shouldForceCompact` flag - Uses `liveUsage` (real-time streaming tokens) with fallback to `lastUsage` - Triggers when remaining tokens ≤ 5200 (2× expected compaction output) - Works even with empty `usageHistory` (first message streaming) ### AIView Integration - New `useEffect` monitors `shouldForceCompact` during active streams - Tracks triggered stream ID to prevent duplicate compactions - Sends compaction request with "Continue with current task" as follow-up ### Queue & Restore Fixes - **Moved restore-to-input** from `stream-abort` handler to IPC `interruptStream` handler - User interrupts (Ctrl+C) still restore queued messages to input - Internal aborts (compaction flow) preserve queue for follow-up ### Shared Constants - `DEFAULT_COMPACTION_WORD_TARGET = 2000` - `WORDS_TO_TOKENS_RATIO = 1.3` - `FORCE_COMPACTION_TOKEN_BUFFER = 5200` (derived: 2 × 2000 × 1.3) ## Testing - 31 unit tests for `checkAutoCompaction` including force compaction scenarios - Manually tested by asking the agent to read a large file in 22 chunks whilst summarising each chunk. The process completed successfully even with a force-compaction triggering mid-task. --- _Generated with `mux`_
github-merge-queue
bot
removed this pull request from the merge queue due to failed status checks
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.
Summary
Automatically triggers compaction when live token usage approaches the context window limit during streaming. This prevents the AI from hitting context limit errors mid-response.
Changes
Force Compaction Logic
checkAutoCompaction()to returnshouldForceCompactflagliveUsage(real-time streaming tokens) with fallback tolastUsageusageHistory(first message streaming)AIView Integration
useEffectmonitorsshouldForceCompactduring active streamsQueue & Restore Fixes
stream-aborthandler to IPCinterruptStreamhandlerShared Constants
DEFAULT_COMPACTION_WORD_TARGET = 2000WORDS_TO_TOKENS_RATIO = 1.3FORCE_COMPACTION_TOKEN_BUFFER = 5200(derived: 2 × 2000 × 1.3)Testing
checkAutoCompactionincluding force compaction scenariosGenerated with
mux