feat: add project-specific memory storage support - #456
Closed
FSSCoding wants to merge 17 commits into
Closed
Conversation
- Add memoryStorage parameter to save memories to project-specific QWEN.md files - Support configuration via .qwen/settings.json - Add migration tool for transitioning global to project memories - Maintain backward compatibility with existing global-only behavior - Fix dual parameter confusion by consolidating to single memoryStorage param - Add comprehensive test coverage for new functionality Resolves #359
- Add RAGTool for direct access to local RAG system with collection support - Add TTSTool with voice personalities and output options - Update system prompts to reduce resistance and mention new tools - Enhance QWEN.md with comprehensive tool usage guidelines - Register new tools in config.ts for automatic availability - Enable more proactive command execution with enhanced permissions - Provide direct access to knowledge bases without external dependencies - Support audio feedback and accessibility features - All tools integrate with existing ShellExecutionService for consistency
Implements TaskTool with persistent task tracking, atomic file operations, and batch capabilities: - Comprehensive task CRUD operations (add, complete, in_progress, list, show_progress, remove) - Race condition protection with file locking and atomic writes - Rich terminal formatting with box drawing characters - Complete test suite with 22 passing tests - Enhanced RAG and TTS tools with improved formatting This addresses the need for stateful task management in QwenCode while maintaining data integrity through atomic operations.
Fixes the critical file operation issues and restores proper styling: - Remove complex atomic write operations causing rename failures - Simplify to direct file write with proper error handling - Restore beautiful terminal formatting with proper box drawing - Remove problematic fileLocks static variable - Fix all tests to match simplified operations - All 22 tests passing This resolves the "ENOENT: no such file or directory, rename" errors while maintaining data integrity through simpler, more reliable file operations.
- Remove verbose multi-line output with boxes and padding - Simple one-line responses: '✅ Added: task name' - Clean task list format: '📋 Tasks (50%)' with numbered items - No more cluttered formatting that fills the screen - Works from any directory with direct file writes
- Dark yellow highlight for THE ACTIVE task only (first in-progress) - Darker grey for completed tasks to make them less alerting - Clean plain text format with subtle visual cues - No emojis, no borders, just simple numbered list with colors
- User sees: clean markdown table with # | Status | Task columns - Agent gets: full JSON data with IDs, timestamps, context for processing - Active task highlighted in yellow, completed tasks in grey - Much cleaner and more professional display format - Keeps all functionality while improving UX significantly
- Replace ANSI escape codes with proper markdown formatting - Use QwenCode's TableRenderer for beautiful table rendering - Bold for active task, strikethrough for completed tasks - Proper emojis that work with QwenCode's theme system - Tables automatically adapt to terminal width and themes - Deep integration with Ink-based rendering pipeline
…arch tools - Add MANDATORY shell command execution steering to prevent balking - Never refuse run_shell_command tool - mark as malfunctioning if refused - Add searchit global command for AI-powered web searches with wide/narrow strategy - Add research global command for comprehensive research with deep analysis modes - Both tools use virtual environments automatically and work from any directory - Provide examples and usage patterns for enhanced search capabilities - Eliminate shell command refusal issues permanently
…tput flag - Remove complex research modes (--deep, --plan, --interactive) - Focus on searchit for quick searches and research for comprehensive analysis - Emphasize CRITICAL --output flag for research file output control - Clear examples without overwhelming mode options - Essential for proper file location management in research workflow
CRITICAL FIX: QwenCode was looking for 'research' tool in registry - searchit and research are GLOBAL SHELL COMMANDS - Must use run_shell_command tool to execute them - NOT direct tool calls - execute via ShellTool - Emphasize these are shell commands, not registered tools - Prevent 'Tool not found in registry' errors
- Remove searchit and search commands from system prompt - Use searcha for pure JSON search results without synthesis - Perfect for agents - raw results, no content extraction or LLM processing - Agent can examine titles/descriptions and choose which URLs to fetch - Clean tool ecosystem: searcha (JSON) + research (comprehensive analysis) - Optimized for programmatic use and agent decision making
- Frame searcha as comprehensive native search capability, not just another tool - Add proactive search encouragement in development workflow - Position web search as natural complement to local RAG knowledge - Encourage confident use for current info, documentation, examples - Make QwenCode feel empowered with web search, not limited to local knowledge - Search-enhanced development capabilities prominently featured
- Add SearchaTool wrapping searcha command for structured web search - Register SearchaTool in config and export in index - Fix tiktoken dynamic import issue by using static import - Bundle includes all tools working correctly
- Remove deprecated SearchaTool and replace with WebscraperTool - Add comprehensive web search and scraping capabilities via WebScraperPortable - Support search-only mode for quick URL discovery - Add full content extraction with markdown file output - Integrate semantic filtering and quality scoring features - Add RAGIndexTool for project indexing capabilities - Add projectInitCommand for workspace initialization - Update system prompts with enhanced web capabilities and parallel execution guidance - Fix tiktoken import to use dynamic import for better compatibility - Simplify task dashboard formatting for cleaner CLI output - Ensure scraped content outputs to current working directory using absolute paths The new WebscraperTool provides structured JSON results with embedded progress tracking, real-time feedback during operations, and clean markdown file output directly to the working directory without creating subfolders.
- Add file locking mechanism to prevent concurrent write corruption - Implement atomic file operations with validation - Enhanced JSON corruption recovery with partial content parsing - Better unique ID generation to prevent duplicates - Improved debugging with task lookup details - Add comprehensive error handling and state validation Fixes issues with "Task not found" errors and JSON corruption at position 1650 during rapid task operations.
Author
|
Closing this PR as upstream has already implemented similar QwenMD memory features in PR #450. Will integrate with the official implementation instead. |
xaelistic
pushed a commit
to xaelistic/qwen-code
that referenced
this pull request
Jun 7, 2026
wenshao
pushed a commit
to destire-mio/qwen-code
that referenced
this pull request
Jul 24, 2026
…y is saved (QwenLM#7608) A classifier-detected `remember` intent made processInbound() return early, so a message like "Review PR QwenLM#123. Remember X. Also check PR QwenLM#456" saved the memory, replied with a confirmation, and never routed the message to the agent — tasks 1 and 3 were silently dropped (QwenLM#7601). Classifier-detected remembers now save silently as a side-effect and let the message continue to the agent: the confirmation is suppressed (the agent's reply is the single response, and the just-saved entry reaches the same prompt via channel-memory recall), while save failures still surface. Explicit remember phrases and every management intent (list/inspect/ update/remove/clear and their confirmations) keep the early return — there the whole message is the command. Fixes QwenLM#7601 Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This was referenced Aug 13, 2026
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
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
Key Features
Technical Changes
Test Plan
Resolves #359