- Improve history ID generation logic to be strictly increasing, preventing "Transition Race" collisions between pending and finalized items.
- Strengthen ID tracking to correctly handle session resumes by ensuring new IDs always start after the highest existing entry.
- Assign unique negative IDs and negatively indexed keys (`pending-${i}`) to pending history items in `MainContent`.
- Add a global `callIdCounter` to `acpClient` to safeguard against sub-millisecond tool call collisions.
- Harden `DenseToolMessage` rendering with stable string keys for conditional UI components.
- Update unit tests to verify strictly increasing ID generation and align with negatively indexed pending IDs.
- Added `isStructuredToolResult` type guard to safely check for `summary` properties on tool results in `ToolResultDisplay`.
- Resolved `data.slice is not a function` errors in UI tests caused by structured tool results falling back to `AnsiOutput` parsing.
- Updated `DiffRenderer` assertions to include `disableColor: false`.
- Updated `ReadManyFilesTool` tests to assert against the new `ReadManyFilesResult` object structure instead of flat strings.
- Removed unsafe type assertions (`any`) in `ToolGroupMessage.compact.test.tsx` and `ToolGroupMessage.tsx` where possible.
- docs: sync settings and configuration with latest schema
- wrap waitUntilReady in `act` to resolve spinner warnings
fix(core): use full file content for accurate diff stats in edit tool
- Ensures the edit tool uses fully calculated new file content (instead of just the replacement snippet) when generating diff stats. This prevents the bug where 'mass deletion' were inaccurately reports in the UI.
- The fix calculates the 'AI-recommended' full file state when user modifications are present. This allows getDiffStat to receive three full-file strings, maintaining accurate telemetry for AI vs. user contributions while fixing the visual bug.
feat(ui): structured compact payloads for list and grep tool output
- leverages structured object return types (ReadManyFilesResult, ListDirectoryResult, GrepResult) to enable rich, expandable compact output in the CLI.
- Updates the underlying core tools (read_many_files, ls, grep) to return these objects and wires up the UI in DenseToolMessage to parse and render them as expandable lists.
- fix(ui): update ToolGroupMessage to support stitched borders and dynamic margins
- fix(ui): implement border stitching in history pushing to eliminate gaps
- test(ui): update snapshots and test assertions for new layout
- style(ui): wrap dense tool payloads with vertical margins
- Adds margins above and below dense payload content to allow compact output to consume a single line until expanded.
- fix(ui): unify spacing logic and border handling for tool groups
- Corrects transitions between compact and standard tools to not add redundant empty lines and ensures history stitching respects boundaries.
- fix(ui): ensure top border is rendered within completed history for standard tool output when following compact tool output
- Addresses an issue where non-compact tools pushed to history at the end of a batch (via onComplete) were missing their top border and proper margin if they followed compact tools already pushed to history.
- The fix updates the onComplete callback in useGeminiStream.ts to be transition-aware. It now explicitly detects when a final push starts with a non-compact tool following a compact tool from the same batch, forcing borderTop: true in that case.
- Previously, the logic relied solely on isFirstToolInGroupRef, which would be false if any earlier tools in the batch had already been pushed, causing the final non-compact tools to incorrectly inherit a borderless state from the preceding compact tools.
----------------------
Note: ToolGroupMessage.tsx and ToolGroupMessage.compact.test.tsx contain 'any' usage/unsafe assertions to be addressed before PR.
- feat(cli): add compactToolOutput ui setting
- feat(cli): add DenseToolMessage component and supporting rendering logic
- feat(cli): implement compact tool output rendering logic in ToolGroupMessage
- feat(core): add tool display name constants and update tool constructors
- feat(cli): update compact output allowlist to use DISPLAY_NAME constants
- test(cli): add compact tool output rendering and spacing tests
-----
Note: Unsafe assignments/assertions/'any' usage still exist in ToolGroupMessage.tsx and DenseToolMessage.tsx and need to be addressed before PR.
Note: Tests contain 'any' casts for mocks that need to be addressed before PR.
