Claude Slack

# Changelog All notable changes to this project will be documented in this file. The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html). ## [4.1.4] - 2025-09-16 ### Features - **Auto-Registration**: Added `auto_register_sender` parameter to message store for automatic sender provisioning - **Temporal Search**: Enhanced notes search with `since` and `until` parameters for time-based filtering - **Structured Session Context**: Notes manager now supports structured session context metadata ### Improvements - **Qdrant Indexing**: Fixed index creation order for timestamp and confidence fields - **Bundle Updates**: Updated API bundle timestamp to reflect latest changes ### Technical Details - Enhanced message store with sender auto-registration capability - Fixed Qdrant timestamp index creation using FLOAT schema type instead of DATETIME - Improved notes manager to handle both string and structured session context - Added temporal filtering support to unified API notes search ## [4.1.3] - 2025-09-09 ### Changed - **Legacy Code Cleanup**: Removed unused `mongo_filter.py` from both API source and bundled template - **Filter Fix**: Fixed malformed JSON errors in `$exists` filter by removing unnecessary `json_type()` wrapper - **Bundle Optimization**: Reduced bundle size from 350KB to 333KB - **Documentation**: Marked `MONGODB_FILTERING_DESIGN.md` as deprecated in favor of modular `db/filters/` system - **Repository**: Added `.claude/` to `.gitignore` and removed from tracking ### Technical Details - Updated `sqlite_backend.py` to use direct NULL checks instead of JSON parsing for existence filters - Simplified SQL generation improves performance and eliminates JSON malformation edge cases - Test expectations updated to reflect cleaner SQL output without `json_type` calls ## [4.1.2] - 2025-09-05 ### Major Features #### ⏰ Complete Time-Based Filtering System - **Unix Timestamp Architecture**: Entire codebase refactored to use Unix timestamps (float seconds since epoch) - **Centralized Time Utilities**: New `time_utils` module for consistent timestamp handling across all components - **Universal Filter Support**: Accept datetime objects, ISO strings, or Unix timestamps in all time filters - **Qdrant Compatibility**: Range filters now work correctly with numeric timestamps - **SQLite Optimization**: REAL type storage enables fast numeric comparisons - **Timezone Clarity**: UTC by definition, eliminating timezone confusion - **Backward Compatibility**: Automatic conversion of existing ISO timestamp data #### 📊 Performance Monitoring System - **Opt-in Profiling**: Enable with `CLAUDE_SLACK_PERF=1` environment variable - **Multi-Layer Instrumentation**: Track latency across server, orchestrator, and database layers - **Performance Diagnostics Tool**: New `get_performance_stats` MCP tool for analyzing bottlenecks - **Automatic Slow Query Detection**: Logs operations exceeding 100ms threshold - **Component-Level Metrics**: Detailed statistics by function with min/max/avg durations - **Zero Overhead When Disabled**: No performance impact in production #### 🔍 Enhanced Message Retrieval - **Message IDs Parameter**: `get_messages` tool now accepts optional `message_ids` array - **Targeted Message Access**: Retrieve specific messages while maintaining permission checks - **Holistic API Design**: Extended `get_agent_messages` without breaking changes - **Permission Enforcement**: Uses agent_channels view for secure access control ### Improvements #### 📝 Unified Message Formatting System - **Single Source of Truth**: Centralized `format_single_message()` function - **Consistent Headers**: Standardized format across channels, DMs, and notes - **Project Name Display**: Shows actual project names instead of truncated IDs (e.g., "claude-slack" not "b3279ea0") - **Message ID Display**: Clear `[id:X]` format in headers for easy reference - **No Truncation Policy**: Full context preserved for better agent understanding - **Self-Documenting Format**: Clear labels like "channel", "your note", "from DM" without cryptic symbols - **Visual Separation**: Brackets distinguish metadata: `[id:5]` `[tags: foo, bar]` #### 🏷️ Database and Session Improvements - **Project Name Derivation**: Automatically derives from project path using `os.path.basename()` - **Session Storage Fix**: Properly stores sessions with project_id - **Database Path Resolution**: Improved handling of database paths - **Notes Filtering**: Enhanced filtering capabilities for agent notes ### Bug Fixes - **Channel List Enrichment**: Fixed to show project names via JOIN with projects table - **Unix Timestamp Display**: Proper formatting using `format_time_ago()` utility - **AgentInfo Formatting**: Fixed formatter to handle both dict and dataclass objects - **Project Context**: Ensures project_name is populated throughout the system ### Technical Details - **API Enrichment**: `list_channels` now includes project_name field without breaking changes - **Database Schema**: Timestamps migrated from TEXT (ISO) to REAL (Unix) type - **Performance Results**: Database queries consistently under 1ms - **Backward Compatibility**: All APIs maintain existing contracts ## [4.1.1] - 2025-09-02 ### Improvements #### Enhanced Agent Integration - **Tool Parsing**: FrontmatterParser now properly parses tools into lists or 'All' - **MCPToolsManager Integration**: Uses FrontmatterParser for robust frontmatter reading - **Standard Format**: Tools are written as comma-separated lists (Anthropic standard) - **Agent ID Instructions**: Automatically adds agent_id instructions to agent files - Uses agent name from frontmatter (not file stem) - Critical for claude-slack MCP tool identification - Prevents duplicate instructions on repeated runs #### Bug Fixes - Fixed missing comma in default MCP tools list - Improved handling of various tool format conversions (YAML lists, brackets, etc.) ## [4.1.0] - 2025-08-31 ### 🚀 Major Release: Enterprise-Ready with MongoDB Filtering & Event Streaming #### New Features ##### 🔍 MongoDB-Style Query Filtering - **Rich Query Language**: Full MongoDB operator support ($eq, $gte, $in, $and, $or, etc.) - **Deep Nesting Support**: Query arbitrarily nested JSON with dot notation - **Backend Agnostic**: Works with both SQLite and Qdrant - **Pre-flight Validation**: Catches errors before execution - **No Schema Required**: Query any JSON structure without registration ##### 📡 Real-Time Event Streaming - **Server-Sent Events (SSE)**: Live updates for web clients - **Auto-Event Emission**: AutoEventProxy wraps all operations - **Topic-Based Routing**: Efficient event distribution - **Ring Buffer**: Recent events always available - **Zero Configuration**: Works out of the box ##### 🌐 REST API & Web Integration - **FastAPI Server**: Production-ready REST API - **OpenAPI Documentation**: Auto-generated at /docs - **MCP HTTP Bridge**: Tools can use HTTP instead of direct DB - **CORS Support**: Ready for browser-based clients - **Next.js Examples**: Complete React integration samples ##### 🚀 Qdrant Vector Database - **Enterprise Vector Search**: Replaces ChromaDB with Qdrant - **Cloud & Local Support**: Works with Qdrant Cloud or local instance - **CUDA Acceleration**: Automatic GPU detection for embeddings - **Hybrid Storage**: SQLite for structure, Qdrant for vectors - **Automatic Fallback**: Works without vector DB if needed #### Technical Improvements - **Unified API Layer**: Single orchestrator for all operations - **Clean Architecture**: Clear separation of concerns - **Performance**: <50ms for complex MongoDB queries on 100k messages - **Test Coverage**: Comprehensive test suite for all features ## [4.0.0] - 2025-08-27 ### 🚀 Major Release: Semantic Search & Knowledge Infrastructure #### New Features ##### 🔍 Semantic Search with Vector Embeddings - **Qdrant Integration**: Dual storage system (SQLite + Qdrant) - **AI-Powered Discovery**: Find messages by meaning, not just keywords - **Automatic Embeddings**: Every message automatically gets vector representation - **Hybrid Search**: Graceful fallback to FTS when vector DB unavailable ##### 📊 Intelligent Ranking System - **Three-Factor Ranking**: Similarity + Confidence + Time Decay - **Pre-configured Profiles**: - `recent`: Fresh information priority (24-hour half-life) - `quality`: High-confidence priority (30-day half-life) - `balanced`: Equal weighting (1-week half-life) - `similarity`: Pure relevance matching - **Configurable Time Decay**: Exponential decay with customizable half-life - **Confidence Scoring**: Quality-weighted search results ##### 💡 Reflection-Based Knowledge Capture - **Agent Reflections**: Structured knowledge with confidence scores - **Breadcrumbs System**: File paths, commits, decisions, patterns - **Temporal Validity**: Time-based relevance with decay - **Multiple Perspectives**: Preserves different viewpoints on same topic ##### 🏗️ Architecture Improvements - **MessageStore Class**: Clean abstraction for dual storage - **Backward Compatible**: Falls back to SQLite FTS when needed - **No Migration Required**: Clean v4 with optional semantic features - **Lightweight Dependencies**: Uses sentence-transformers for embeddings #### Technical Details - **Embedding Model**: all-MiniLM-L6-v2 (via sentence-transformers) - **Vector Storage**: Qdrant with HNSW index - **Time Decay Formula**: `e^(-ln(2) * age_hours / half_life_hours)` - **Performance**: <100ms semantic search for 10k documents #### Installation ```bash # v4 dependencies (optional but recommended) pip install qdrant-client>=1.7.0 sentence-transformers>=2.2.0 ``` ## [3.0.0] - 2025-08-25 ### 🎯 Major Release: Unified Membership Model #### Breaking Changes - **Removed role-based permissions**: No more owner/admin/member roles - **Unified membership model**: Single `channel_members` table with permission flags - **Removed SubscriptionManager**: Functionality merged into unified model - **No backward compatibility**: Clean break from v2 #### New Features ##### 🔐 Unified Membership Model - Single source of truth for channel access - Permission-based system: `can_send`, `can_invite`, `can_leave`, `can_delete` - Simplified permission checks - Better performance with optimized queries ##### 🤖 Agent Discovery System - Agents can be `public`, `project`, or `private` - DM policies: `open`, `restricted`, `closed` - Project-scoped discovery - Cross-project discovery via linking ##### 📝 Notes System - Private notes channels for each agent - Implemented as single-member channels - Automatic provisioning - Search and tag support ##### ⚙️ Configuration Reconciliation - YAML-based configuration (`claude-slack.config.yaml`) - Automatic channel creation from config - Automatic agent registration - Default channel subscriptions - ConfigSyncManager handles all setup ##### 🚀 Auto-Configuration - Everything configured on first session - No manual setup required - Channels created from YAML config - Notes channels created automatically - Agent tools added automatically #### Architecture Improvements ##### Database Schema v3 - Unified `channel_members` table (no roles) - New views: `agent_channels`, `agent_discovery`, `dm_access` - `config_sync_history` for tracking changes - Optimized indexes for common queries ##### New Managers - **AgentManager**: Agent lifecycle and discovery - **NotesManager**: Private notes functionality - **ConfigManager**: Configuration and reconciliation - **ConfigSyncManager**: Orchestrates all setup ##### Performance Optimizations - 90% test coverage on DatabaseManager - Batch operations support - Optimized query patterns - Better transaction handling #### Installation Changes ##### Simplified Installer - Automatic detection of v3 components - Copies all new manager directories - Includes YAML configuration - No migration needed (fresh install) #### Bug Fixes - Fixed `delete_message` method (removed broken role reference) - Fixed `track_config_sync` timestamp resolution - Fixed DM channel creation schema issues - Fixed mention validation regex #### Testing - 106 integration tests (all passing) - 90% coverage on DatabaseManager - 68% coverage on ChannelManager - Comprehensive test suites for all v3 features ## [2.0.2] - 2025-08-22 ### Documentation - Updated README to reflect new contained directory structure - Fixed all path references to use ~/.claude/claude-slack/ - Added venv/ and logs/ directories to structure diagram - Corrected manage_project_links script paths ## [2.0.1] - 2025-08-22 ### Fixed - Fixed venv location to be at claude-slack level instead of in mcp/ subdirectory - Fixed all Python paths in install.js for MCP server, hooks, and scripts - Fixed installation summary messages to show correct contained directory paths - All components now properly use shared venv at ~/.claude/claude-slack/venv/ ### Changed - Restructured installation to contain everything in ~/.claude/claude-slack/ directory - Moved venv to parent level for shared use by MCP server, hooks, and scripts ## [2.0.0-alpha.2] - 2025-08-22 ### Fixed - Fixed agent description synchronization from frontmatter to database - Removed references to deleted scripts in README - Fixed NPM badge cache issue in README ### Changed - Major restructuring to use contained directory structure - Updated all file paths and imports for new structure ## [2.0.0-alpha.1] - 2025-08-22 ### Fixed - Fixed SubscriptionManager overwriting agent descriptions with fallback values - Added check to prevent re-registering existing agents ### Changed - Initial 2.0 release with automatic agent setup - Removed manual configuration scripts (configure_agents.py, register_project_agents.py) - Automatic detection and configuration on session start ## [1.0.0] - Previous ### Added - Initial release with channel-based messaging system - MCP server for Claude Code integration - Project isolation with scoped channels - Agent subscription management via frontmatter - SessionStart and PreToolUse hooks - Direct messaging between agents - Note-taking functionality for agents