release/v6.0.2

[code-crusher]

• release/v5.7.1 (release/v5.7.1 #73)
• release/v5.7.2 (release/v5.7.2 #74)
• update readme (update readme #75)
• release/v5.7.3 (release/v5.7.3 #76)
• release/v5.7.6 (release/v5.7.6 #77)
• release/v5.7.7 (release/v5.7.7 #78)
• release/v6.0.0 (release/v6.0.0 #79)
• multi file edits + UI cleanups
• feat: update condense, task, assistant-message, and exploration UI components
• chore: remove .playwright-mcp from tracking and add to gitignore
• feat: enhance slash commands, chat UI, and add custom icons
• fix parallel tool calls exec
• fix filewrite tool response
• fix todo lists
• update pkg version
• create file in file change tracker
• optimise for light theme

[matterai-app]

Context

This PR introduces the multi_file_edit tool to allow the AI agent to batch multiple file modifications into a single operation, significantly improving efficiency. It also adds a /compact slash command and automated context condensation logic to manage token limits proactively.

Implementation

Summary By MatterAI

🔄 What Changed

• New Tool: Added multi_file_edit for batching text replacements across multiple files.
• Slash Command: Introduced /compact (alias for /condense) to manually trigger context window reduction.
• Context Management: Implemented checkAndCondenseContext in Task.ts to automatically summarize history when token usage exceeds thresholds (default 80%).
• Parser Robustness: Enhanced AssistantMessageParser to handle natural language text inside tool arguments during streaming and fix double-encoded parameters.
• UI/UX: Added slash command chips, improved reasoning block scrolling, and integrated VS Code theme-aware shadows for sticky messages.

🔍 Impact of the Change

• Performance: Reduces the number of round-trips to the LLM by batching edits.
• Reliability: Prevents context window overflows by condensing history before executing heavy tools like read_file.
• User Experience: Provides smoother streaming feedback and better visual integration with VS Code themes.

📁 Total Files Changed

Click to Expand

File	ChangeLog
multi_file_edit.ts	New Tool implementation for batch file editing logic.
Task.ts	Context Logic added for auto-condensation and /compact command handling.
AssistantMessageParser.ts	Parser Fix for streaming JSON arguments and double-encoding.
presentAssistantMessage.ts	Orchestration Update to support multi-file edits and pre-tool condensation.
system.ts	Prompt Update adding instructions for multi_file_edit usage.
ChatTextArea.tsx	UI Enhancement for rendering slash command chips and caret management.
fileEditTool.ts	Defensive Fallback to redirect edits arrays to the multi-edit tool.
kilocode-models.ts	Model Cleanup removing deprecated Axon model definitions.

🧪 Test Added/Recommended

Added

• AssistantMessageParser.spec.ts: Tests for multi-tool responses and natural language in streaming arguments.
• presentAssistantMessage.spec.ts: Verifies execution of multiple tool blocks in a single response.

Recommended

• Concurrency Tests: Verify multi_file_edit behavior when two edits target overlapping lines in the same file.
• Threshold Tests: Validate checkAndCondenseContext triggers correctly at various autoCondenseContextPercent levels.

🔒 Security Vulnerabilities

• N/A: No new security vulnerabilities detected. Access control is maintained via existing rooIgnoreController.

Screenshots

before	after
Standard text slash commands	Visual chips for /compact and /newtask

How to Test

1. Start a task and perform 2+ file edits; verify the agent uses multi_file_edit instead of multiple file_edit calls.
2. Type /compact in the chat to manually trigger context condensation.
3. Fill the context window near the limit and trigger a read_file; verify the agent auto-condenses before reading.

Get in Touch

• Handle: @matterai_agent

[matterai-app]

🧪 PR Review is completed: Solid release PR introducing multi-file edit, context condensation, and native tool call streaming fixes. Found a few runtime safety and async handling issues in the new Task and executeCommandTool code.

Skipped files

• webview-ui/src/i18n/locales/en/chat.json: Skipped file pattern

⬇️ Low Priority Suggestions (1)

src/core/tools/executeCommandTool.ts (1 suggestion)

Location: src/core/tools/executeCommandTool.ts (Lines 221-221)

🟠 Logic Error

Issue: The new task.say("command_output", ...) call is missing await, while all other task.say calls in this file are awaited. If say rejects or the caller relies on the message being posted before completed = true, this creates a race condition or unhandled rejection risk.

Fix: Await the task.say call to maintain consistency with the surrounding async flow.

Impact: Eliminates potential unhandled promise rejections and ensures correct message ordering

-  			task.say("command_output", result, undefined, undefined, undefined, undefined, { isNonInteractive: true })
+  			await task.say("command_output", result, undefined, undefined, undefined, undefined, { isNonInteractive: true })
+  

[matterai-app]

🔴 Logic Error

Issue: contextWindow can be undefined or 0 for some models, causing contextPercent to evaluate as NaN or Infinity. The comparison contextPercent < effectiveThreshold becomes unpredictable and may trigger or skip condensation incorrectly.

Fix: Guard against invalid contextWindow before calculating the percentage.

Impact: Prevents erroneous condensation behavior when model metadata is incomplete

Suggested change

	const contextWindow = modelInfo.contextWindow
	// Calculate context usage percentage
	const contextPercent = (100 * contextTokens) / contextWindow
	const contextWindow = modelInfo.contextWindow
	// Guard against invalid context window to prevent NaN/Infinity
	if (!contextWindow) {
	return false
	}
	// Calculate context usage percentage
	const contextPercent = (100 * contextTokens) / contextWindow

[matterai-app]

🟡 Code Quality

Issue: Debug console.log statement left in production code can leak internal state (token counts, thresholds) and clutter extension host logs.

Fix: Remove the debug log or replace it with a proper telemetry/verbose logger.

Impact: Keeps production logs clean

Suggested change

	console.log(
	`[Task#${this.taskId}] Pre-tool context check: ${contextTokens} tokens (${contextPercent.toFixed(1)}%) ` +
	`exceeds threshold ${effectiveThreshold}%. Triggering condensation.`,
	)