🤖 feat: add auto-compaction configuration UI #685
Merged
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
This was referenced Nov 20, 2025
github-merge-queue bot
pushed a commit
that referenced
this pull request
Nov 21, 2025
## Stack 1. #685 1. #683 1. #670 1. #650 ⬅ This PR (base) ## Problem Compact continue messages were handled by a frontend hook that watched workspace states and manually sent continue messages after compaction. This was complex, had potential race conditions, and poor separation of concerns. Relates to #651. ## Solution Use the existing message queue system: - Backend queues continue message when compaction starts - Queue auto-sends when stream ends (existing behavior) - Clear queue on error paths **Benefits:** Simpler (-134 lines), more reliable, better UX (continue message visible in queue). _Generated with `mux`_
github-merge-queue bot
pushed a commit
that referenced
this pull request
Nov 21, 2025
_Generated with `mux`_ ## Stack 1. #685 1. #683 1. #670 ⬅ This PR 1. #650 (base) ## Summary Moves history compaction handling from WorkspaceStore (frontend) to agentSession (backend) to centralize server-side operations and fix race conditions. Relates to #651. ## Changes ### Backend (agentSession.ts) - Added `handleCompactionCompletion()` - detects compaction stream-end, extracts summary from event.parts, performs history replacement - Added `handleCompactionAbort()` - handles Ctrl+A (accept early with `[truncated]`) and Ctrl+C (cancel) flows - Added `performCompaction()` - atomically replaces chat history with summary message including cumulative usage - Implemented `abandonPartial` flag flow from IPC through to StreamAbortEvent - Extracts truncated message content from history instead of partialService ### Frontend (WorkspaceStore.ts) - Removed `handleCompactionCompletion()` and `handleCompactionAbort()` methods - Removed `performCompaction()` method - Removed `processedCompactionRequestIds` Set - Simplified `cancelCompaction()` - just calls `interruptStream` with `abandonPartial: true` - Fixed Ctrl+A keybind to pass `abandonPartial: false` for early accept ### Shared - Updated `StreamAbortEvent` to include `abandonPartial?: boolean` - `historyService.clearHistory()` now returns deleted sequence numbers - Created `calculateCumulativeUsage()` utility in `displayUsage.ts` to extract and sum usage from messages ## Testing - [x] Manual: `/compact` completes successfully - [x] Manual: Ctrl+A during compaction accepts early with `[truncated]` - [x] Manual: Ctrl+C during compaction cancels and enters edit mode - [x] Verify cumulative usage preserved across multiple compactions
ethanndickson
force-pushed
the
frontend-auto-compaction-queue
branch
from
2438689 to
087e85f
Compare
ethanndickson
force-pushed
the
auto-compaction-config
branch
from
52d11bd to
0a73025
Compare
ethanndickson
force-pushed
the
auto-compaction-config
branch
2 times, most recently
from
7faf26f to
d9e9139
Compare
Member
Author
|
@codex review |
|
Codex Review: Didn't find any major issues. Breezy! 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 đź‘Ť. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
ethanndickson
force-pushed
the
auto-compaction-config
branch
from
d9e9139 to
84b2830
Compare
- Countdown warning (60-69%): Small grey text, right-aligned - Urgent warning (70%+): Keep prominent blue box styling - Makes countdown unobtrusive while keeping urgent warning visible _Generated with `mux`_
ethanndickson
force-pushed
the
frontend-auto-compaction-queue
branch
from
e2149ae to
444c3ba
Compare
ethanndickson
force-pushed
the
auto-compaction-config
branch
from
2ddd9fb to
803b0b3
Compare
Adds !isCompacting check to shouldAutoCompact calculation to prevent queueing a second compaction request when user sends a message while the first compaction is still running. Without this check, messages sent during compaction would trigger another compaction, resulting in back-to-back compactions and delayed user messages. _Generated with `mux`_
Adds 23 unit tests covering: - Basic functionality (safe defaults, threshold detection) - Usage calculation (last entry vs cumulative, historical usage handling) - 1M context mode (model support, fallback behavior) - Edge cases (zero tokens, custom thresholds, boundary conditions) - Percentage calculation accuracy All tests verify the infinite loop fix - that historical usage from compaction is correctly excluded from threshold calculations. Tests run: 23 pass, 0 fail, 62 expect() calls _Generated with `mux`_
- Add configurable auto-compaction threshold (50-90%) per workspace - Extract threshold constants to ui.ts (DRY) - Create reusable useClampedNumberInput hook for numeric inputs - Add settings to right sidebar with checkbox and percentage input - Wire settings through to shouldAutoCompact check - Use existing HelpIndicator pattern for tooltips
- Rename shouldAutoCompact to checkAutoCompaction in tests - Add enabled parameter to all test function calls - Remove enabled field assertions (no longer in return type) - Fix function parameter ordering for custom threshold tests - All 23 tests passing
ethanndickson
force-pushed
the
auto-compaction-config
branch
from
803b0b3 to
71eb2b2
Compare
ethanndickson
force-pushed
the
frontend-auto-compaction-queue
branch
2 times, most recently
from
a4f8272 to
b184dcf
Compare
…into auto-compaction-config
Member
Author
|
@codex review |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
Member
Author
|
@codex review |
|
Codex Review: Didn't find any major issues. Nice work! 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 đź‘Ť. 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
Nov 25, 2025
## Stack 1. #685 1. #683 ⬅ This PR 1. #670 1. #650 (base) ## Summary Adds automatic context compaction that triggers at 70% usage, with progressive countdown warnings starting at 60%. <img width="905" height="155" alt="image" src="https://github.com/user-attachments/assets/b0db20c5-c377-44bb-891c-f8ddadd561c8" /> <img width="891" height="194" alt="image" src="https://github.com/user-attachments/assets/6385cfd2-5e3c-45ec-afce-935dae56ad1a" /> Relates to #651. ## Key Changes **Auto-Compaction:** - Triggers automatically when current context usage reaches 70% of model's context window - Queues user's message to send after compaction completes - Includes image parts in continue messages **Progressive Warnings:** - Shows countdown at 60-69% usage: "Context left until Auto-Compact: X% remaining" - Shows urgent message at 70%+: "⚠️ Approaching context limit. Next message will trigger auto-compaction." **Implementation:** - New `shouldAutoCompact()` utility centralizes threshold logic with configurable constants - Returns `{ shouldShowWarning, usagePercentage, thresholdPercentage }` - Uses **last usage entry** (current context size) to match UI token meter display - Excludes historical usage from threshold check to prevent infinite compaction loops - `ContinueMessage` type now includes optional `imageParts` ## Technical Details **Usage Calculation:** The auto-compaction check uses the most recent usage entry from `usageHistory` to calculate the current context size. This matches the percentage displayed in the UI token meter and correctly handles post-compaction scenarios: - **Before compaction**: Last entry represents full context → triggers at 70% correctly - **After compaction**: Last entry excludes historical usage → resets to actual context size - **Historical usage preserved**: Remains in usage history for cost tracking, but not used for threshold calculations This prevents the infinite loop where post-compaction workspaces would continuously re-compact because historical usage tokens were being included in the threshold check. ## Future Work Future PRs will add user settings to configure auto-compaction (enable/disable, custom threshold). _Generated with `mux`_
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.
Stack
Relates to #651.
Adds per-workspace settings for auto-compaction (any percentage between 50 and 90):
New Features:
Implementation:
ui.tsfor DRYuseClampedNumberInputhook for numeric input validationshouldAutoCompactto accept settings parametersHelpIndicatorfor tooltips)Settings are forked with workspace and cleaned up on workspace deletion.
Generated with
mux