- Two-Layer Intelligence: Streamlined from three layers to two for better maintainability while preserving all capabilities
- LAYER 1: SERENA SEMANTIC INTELLIGENCE - Semantic code navigation, symbol analysis, project memory, and precise code targeting
- LAYER 2: SEQUENTIAL THINKING COMPLEX REASONING - Multi-step problem decomposition, hypothesis development, and systematic solution synthesis
- Removed Zen MCP Dependency: Eliminated external Zen MCP dependency, simplifying architecture and reducing complexity
- DRY Principles Integration: Moved DRY (Don't Repeat Yourself) principles from shared files into global/CLAUDE.md
- Complexity-Based Triggers: Integrated automatic activation rules directly into global instructions
- Streamlined Workflows: Simplified intelligence flows while maintaining comprehensive problem-solving capabilities
- Pattern Reuse Enforcement: Systematic search and reuse before generating new code
- All 9 Agents Updated: Refactored all agents to use two-layer intelligence architecture
- Removed Shared Directory: Eliminated
/agents/shared/directory by incorporating content into global instructions - Simplified Orchestration: Streamlined agent workflows with focus on Serena + Sequential Thinking
- Maintained PROACTIVE Behaviors: Preserved all aggressive excellence standards and automatic intervention patterns
- Reduced Complexity: Simpler architecture with two layers instead of three
- Better Documentation: Clearer structure with consolidated instructions
- Easier Updates: Single source of truth for global patterns and principles
- Preserved Capabilities: All agent capabilities maintained with simpler implementation
- Zen MCP References Removed: All Zen MCP tool references eliminated from agent files
- Shared Files Removed:
/agents/shared/directory no longer exists - Intelligence Layer Change: Updated from three-layer to two-layer architecture
- Updated all agent files (9 total) to reflect two-layer architecture
- Enhanced global/CLAUDE.md with comprehensive DRY principles and trigger patterns
- Updated main CLAUDE.md and README.md to v1.5
- Revised capabilities summary to reflect streamlined architecture
All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- LiveView Streams: Complete integration of Phoenix 1.8 stream functionality (
stream/3,stream_insert/3,stream_delete/2) - Modern Navigation: Updated navigation patterns using
<.link navigate={path}>andpush_navigate/2 - Verified Routes: Full support for
~psigil for compile-time route checking - Enhanced Presence: Advanced Phoenix Presence with
fetch/2andhandle_metas/4callbacks
- 30+ Semantic Components: Complete expertise in DaisyUI component library (buttons, cards, forms, navigation, modals, alerts, badges)
- Theme Management: Advanced theming system with 30+ built-in themes and dark mode support
- Responsive Design: Mobile-first responsive patterns with Tailwind CSS utility classes
- Accessibility Features: ARIA compliance and keyboard navigation support built into components
- backend-developer: Added comprehensive Phoenix 1.8 expertise with LiveView Streams, modern navigation, verified routes, and enhanced Presence patterns
- frontend-developer: Enhanced with complete DaisyUI component system expertise, theme management, and Phoenix LiveView integration patterns
- Shared Patterns: Updated
mcp-orchestration.mdwith Phoenix 1.8 + DaisyUI integration workflows and intelligence cascades
- Real-time UI Updates: Phoenix 1.8 streams combined with DaisyUI components for dynamic interfaces
- Modern Navigation: Phoenix 1.8 verified routes with DaisyUI navigation components
- Theme Integration: JavaScript hooks for DaisyUI theme persistence and system preferences
- Component Integration: Seamless integration between Phoenix LiveView and DaisyUI component library
- Enhanced agent documentation with Phoenix 1.8 and DaisyUI expertise
- Updated shared orchestration patterns for modern Phoenix development
- Added integration examples and best practice patterns
- Comprehensive component library documentation
- LAYER 1: SERENA SEMANTIC INTELLIGENCE - Semantic code navigation (
find_symbol,get_symbols_overview,search_for_pattern,find_referencing_symbols) - LAYER 2: ZEN AI-POWERED ANALYSIS - 16 sophisticated tools including
thinkdeep,analyze,debug,refactor,consensus,planner,secaudit,testgen,codereview,precommit,tracer,docgen,chat,challenge, andsequential-thinking - LAYER 3: SEQUENTIAL THINKING COMPLEX REASONING - Multi-step problem decomposition, hypothesis development, and systematic solution synthesis
- 5 Pre-defined Workflows: Advanced debugging, architectural decisions, security hardening, performance optimization, and code quality excellence
- Semantic-First Analysis: Always understand code context through Serena before any analysis or modification
- Multi-Model Consensus Validation: Critical decisions validated by 2-3 specialized AI models from 28+ available models
- Assumption Challenge Workflows: Systematic validation using Zen's challenge tool to prevent confirmation bias
- Context-Aware Orchestration: Intelligent model selection based on problem type and complexity
- Deep Analysis:
gemini-2.5-pro(1M context) +anthropic/claude-opus-4.1+openai/o3consensus - Security Analysis:
anthropic/claude-opus-4.1+deepseek/deepseek-r1-0528(thinking mode) +openai/o3validation - Performance Optimization:
openai/o3+gemini-2.5-flash+anthropic/claude-opus-4.1validation - Complex Reasoning:
deepseek/deepseek-r1-0528+gemini-2.5-profor advanced reasoning scenarios
- architect: Revolutionary three-layer orchestration with system-wide analysis cascades and architectural decision validation
- backend-developer: Sophisticated Elixir/Phoenix/OTP intelligence with Ash framework migration patterns and complex bug investigation
- All Agents: Enhanced with three-layer intelligence capabilities (7 remaining agents planned for future releases)
/agents/shared/mcp-orchestration.md: Revolutionary three-layer intelligence orchestration patterns with 16-tool Zen workflows/agents/shared/trigger-patterns.md: Enhanced sophisticated trigger patterns for advanced reasoning activation/global/CLAUDE.md: Enhanced with comprehensive three-layer intelligence architecture guidelines
- 10x Analysis Depth: Systematic use of all available semantic, AI, and reasoning tools
- Zero Blind Spots: Complete code understanding through three-layer intelligence architecture
- Validated Decisions: All critical choices validated by multiple specialized AI models
- Assumption Testing: Systematic challenge of approaches to prevent bias and errors
- Adaptive Intelligence: Context-aware selection of analysis tools and AI models
- Failure-Resilient Patterns: Robust recovery and fallback strategies for interrupted workflows
- Updated main CLAUDE.md with comprehensive v1.4.0 three-layer intelligence architecture
- Enhanced project overview with unprecedented code intelligence capabilities summary
- Updated all version references throughout documentation to v1.4.0
- Added revolutionary intelligence loop documentation and advanced problem-solving workflows
- Global Instructions File: Added
/global/CLAUDE.mdwith universal Claude Code best practices - Task Delegation Framework: Mandatory agent delegation for all complex tasks with parallel execution support
- Enhanced Git Workflows: Automatic reviewer declarations using
git configand standardized Claude Code attribution - Advanced Code Analysis: Integrated ast-grep for syntax-aware structural code searches over text-based searches
- Global-Local Coordination: Universal standards seamlessly integrated with project-specific PROACTIVE agent behaviors
- Universal Task Management: All agents now follow mandatory delegation patterns with agent-first approach
- Parallel Agent Execution: Multi-faceted problems automatically trigger coordinated parallel agent workflows
- Syntax-Aware Analysis: Language understanding integrated into all agent search and analysis workflows
- Quality Gate Integration: Git workflows now enforce combined global standards and agent quality requirements
- Unified Excellence: Global Claude Code standards merged with domain-specific PROACTIVE and AGGRESSIVE behaviors
- Project Structure: Added
/global/directory to project documentation and structure - Integration Documentation: Comprehensive global-local coordination patterns and workflows
- Version Updates: Updated CLAUDE.md, README.md, and VERSION file to reflect v1.3.2
- Enhanced Features: Detailed integration between universal instructions and specialized agent capabilities
- Removed Standalone Memory MCP: Eliminated separate Memory MCP service
- Serena MCP Integration: Memory functionality now fully integrated into Serena MCP
- Enhanced Memory Tools: All agents now use
write_memoryandread_memorytools for project knowledge persistence - Simplified Architecture: Reduced complexity by consolidating memory management into existing Serena MCP
- Improved Documentation: Updated all agent documentation to reflect proper Serena MCP memory capabilities
- Consistent Terminology: Aligned all references to use "Serena MCP - Project Memory" with actual tool functionality
- Updated all 9 agent files with correct Serena MCP memory references
- Removed
rules/memory.md(no longer needed) - Updated CLAUDE.md, README.md, and CHANGELOG.md to reflect memory integration
- Corrected terminology from "Memory MCP" to proper Serena MCP memory tools
- Surface + HEEx Support: All agents now expert in both Surface (.sface) and HEEx (.heex) templating
- Project-Based Standards: Agents choose appropriate templating approach based on project conventions
- Enhanced Specialization: Core agents enhanced with domain-specific expertise while maintaining consistent tooling
- Non-Prescriptive Approach: Agents adapt to existing project patterns rather than enforcing specific engines
- backend-developer: Enhanced Ash Framework 3.5+ and Elixir OTP expertise with flexible templating
- frontend-developer: Svelte 5 runes mastery + Surface/HEEx dual expertise
- api-specialist: HubSpot API integration patterns with comprehensive OAuth 2.0 flows
- quality-specialist: JWT/JOSE security patterns and AshCloak encryption expertise
- architect: PlantUML diagramming and multi-system architecture patterns
- Removed Legacy References: Cleaned up outdated references and consolidated documentation
- Streamlined Structure: Eliminated redundant specialized agent files
- Enhanced Documentation: Updated all version references and capability descriptions
- Sequential Thinking Integration: Automatic activation for complex problems (≥3 steps)
- Zen MCP Orchestration: Systematic workflows combining multiple Zen tools
- Multi-Model Consensus: Critical decisions validated by multiple AI models
- Automated Trigger Patterns: Complexity-based activation thresholds
- Cascading Workflows: Orchestrated sequences (debug → analyze → refactor → consensus)
- Cross-Agent Coordination: Systematic collaboration patterns for complex features
/agents/shared/mcp-orchestration.md: Automated MCP workflow integration patterns/agents/shared/trigger-patterns.md: Systematic trigger patterns for advanced reasoning- Enhanced
/agents/shared/dry-principles.md: Added MCP integration for DRY enforcement
- All 9 Agents: Mandatory integration with shared frameworks (DRY + MCP Orchestration + Trigger Patterns)
- Sequential Thinking: Automatic problem decomposition for debugging, optimization, architecture
- Zen MCP Tools: Comprehensive tool coverage including zen_debug, zen_secaudit, zen_consensus, zen_planner, zen_tracer
- Workflow Automation: Context-aware model selection and failure handling patterns
- Quality Metrics: Workflow effectiveness tracking and continuous improvement
- architect: Enhanced architectural reasoning with zen_consensus and sequential planning
- backend-developer: Systematic debugging workflows and performance optimization cascades
- quality-specialist: Orchestrated security auditing and performance analysis workflows
- All Agents: Expanded Zen tool coverage, automated trigger evaluation, workflow orchestration
- Complexity Detection: Automatic assessment of task complexity against trigger thresholds
- Model Orchestration: Context-aware selection of optimal AI models for different tasks
- Failure Resilience: Comprehensive fallback strategies for MCP unavailability
- Continuous Learning: Pattern recognition and workflow optimization based on success metrics
- Shared Patterns Library: Created
/agents/shared/dry-principles.mdwith core DRY directives - Pattern Recognition: All agents now search for existing patterns before generating code
- Code Reuse Enforcement: Systematic extraction and reuse of common functionality
- Component Composition: Build from smaller, reusable parts instead of duplication
- All 9 Agents: Added mandatory DRY principle enforcement sections
- backend-developer: Elixir-specific patterns for modules, error handling, and changesets
- frontend-developer: Component composition, design tokens, and custom hooks
- architect: System-level DRY enforcement and architectural patterns
- data-engineer: Schema reuse, query patterns, and ETL pipeline components
- api-specialist: Endpoint patterns, response formats, and validation rules
- test-engineer: Test fixtures, utilities, and factory patterns
- quality-specialist: Security patterns, performance optimizations, audit templates
- product-coordinator: Documentation templates and process patterns
- devops-engineer: Infrastructure modules, CI/CD templates, and configuration management
- Updated installation instructions to include
/agents/shared/directory - Added DRY principles section to project overview
- Updated version to 1.1.1 across all documentation
- Enhanced README with new features and installation steps
- Reduced code duplication across projects
- Improved maintainability through pattern extraction
- Enhanced code quality with systematic reuse
- Better architectural consistency
- SERENA MCP: All agents now use semantic code intelligence for navigation and refactoring
- ZEN MCP: Integrated advanced AI analysis tools (thinkdeep, consensus, refactor, debug, etc.)
- TIDEWAVE MCP: Backend and test agents leverage Elixir-specific testing and code generation
- SPECIALIZED MCPs: Each agent uses domain-specific MCPs for enhanced capabilities
- Figma MCP for frontend design integration
- Browser MCP for UI/E2E testing
- Notion MCP for documentation management
- Serena MCP for project knowledge persistence through memory tools (write_memory/read_memory)
- Context7 MCP for up-to-date library documentation
- Brave Search MCP for real-time best practices
- Sequential Thinking MCP for complex problem decomposition
- Minimax MCP for media generation when needed
- architect: Now uses Zen MCP for deep architectural analysis and consensus building
- backend-developer: Integrated Tidewave MCP for Elixir test generation and scaffolding
- frontend-developer: Leverages Figma MCP for design sync and Browser MCP for testing
- data-engineer: Enhanced with semantic code analysis and query optimization MCPs
- api-specialist: Uses Notion MCP for comprehensive API documentation
- test-engineer: Integrated Tidewave and Browser MCPs for comprehensive testing
- quality-specialist: Enhanced security auditing with specialized MCPs
- product-coordinator: Primary use of Notion MCP for project management
- devops-engineer: Monitoring and infrastructure MCPs for better automation
- Updated CLAUDE.md with MCP integration details
- Enhanced README.md with MCP features and capabilities
- Added MCP integration sections to all 9 agent definitions
- Updated project structure documentation
- Agents now understand code semantically, not just text patterns
- Multi-model consensus for critical architectural decisions
- Real-time documentation access without training data limitations
- Persistent project knowledge across sessions
- Framework-specific code generation and testing
- Assigned unique distinct colors to all 9 agents for better visual identification
- Improved color separation across the spectrum to avoid similar colors
- Each agent now has a clearly distinguishable color
- Updated agent name fields to match filenames exactly (e.g.,
name: frontend-developerinstead ofname: Frontend Developer) - Ensures consistency between agent file naming and internal name identifiers
- MASSIVE CONSOLIDATION: Reduced from 23 agents to 9 elite agents (61% reduction)
- PROACTIVE AGENTS: All agents now automatically detect and fix issues without permission
- AGGRESSIVE EXCELLENCE: Zero tolerance for suboptimal code or technical debt
- COMPREHENSIVE SOLUTIONS: Every fix exceeds requirements and implements best practices
- Strategic vision, system design, and comprehensive planning
- AGGRESSIVELY challenges architectural decisions and forces optimal solutions
- AUTOMATICALLY creates PRDs, technical specs, and documentation
- ZERO TOLERANCE for non-Ash patterns - automatically converts Ecto to Ash
- PROACTIVELY optimizes performance and implements proper error handling
- AGGRESSIVELY enforces OTP principles and LiveView best practices
frontend-developer (Consolidates: ux-designer + web-developer + sveltekit-developer + flutter-developer + reactnative-developer)
- IMMEDIATELY fixes UI/UX inconsistencies and accessibility issues
- AGGRESSIVELY enforces design systems across ALL platforms
- AUTOMATICALLY optimizes performance and user experience
- ZERO TOLERANCE for inefficient queries - automatically optimizes and indexes
- PROACTIVELY designs optimal schemas and implements proper migrations
- AGGRESSIVELY enforces data integrity and implements RDF/SPARQL when needed
- IMMEDIATELY fixes API inconsistencies and implements OpenAPI specifications
- PROACTIVELY enforces REST principles, authentication, and rate limiting
- AUTOMATICALLY generates complete API documentation and integration guides
- ZERO TOLERANCE for untested code - automatically generates comprehensive test suites
- AGGRESSIVELY enforces 100% code coverage and quality metrics
- AUTOMATICALLY sets up CI/CD testing pipelines and quality gates
- IMMEDIATELY fixes security vulnerabilities and performance bottlenecks
- PROACTIVELY implements OWASP best practices and proper authentication
- AUTOMATICALLY conducts security audits and performance profiling
- PROACTIVELY identifies project risks and implements mitigation strategies
- AGGRESSIVELY enforces Agile best practices and proper sprint planning
- AUTOMATICALLY creates comprehensive project documentation and tracking
- ZERO TOLERANCE for manual deployments - automatically implements full CI/CD
- AGGRESSIVELY enforces Infrastructure as Code and proper security practices
- AUTOMATICALLY implements backup strategies, disaster recovery, and scaling
/agents/- Elite 9 PROACTIVE and AGGRESSIVE agents/super-agents/- Advanced multi-agent coordination patterns- Enhanced
/rules/with PROACTIVE behavior enforcement
- ZERO PERMISSION REQUIRED - Agents fix problems immediately upon detection
- COMPREHENSIVE SOLUTIONS - Never implement partial fixes, always exceed requirements
- AGGRESSIVE ENFORCEMENT - Zero tolerance for suboptimal code or technical debt
- AUTOMATIC INTERVENTION - Proactively detect and resolve issues without being asked
- EXCELLENCE BY DEFAULT - Every solution must exceed industry best practices
- CLAUDE.md completely rewritten to reflect PROACTIVE architecture
- README.md transformed with revolutionary messaging and elite agent descriptions
- CHANGELOG.md enhanced with comprehensive v1.0.0 breakthrough documentation
- All agent descriptions emphasize PROACTIVE and AGGRESSIVE behaviors
- Agent names completely restructured (23 → 9 agents)
- All agents now operate in PROACTIVE mode by default
- Legacy collaborative patterns replaced with automatic coordination
- Enhanced framework rules with zero-tolerance enforcement
- Renamed
code-fixertocode-reviewerfor better clarity of purpose - Updated all documentation references
- Renamed agents for brevity and clarity:
code-quality-analyzer.md→code-fixer.mdsemantic-data-specialist.md→ontology-specialist.md
- Enhanced all agents with "ALWAYS use this agent when..." directives for better proactive invocation
- Added specific proactive triggers to 15 key agents to ensure automatic usage in relevant contexts
- Updated all documentation to reflect new agent names
- Agent proactivity: All agents now have clear triggers for when they should be automatically invoked
- Documentation consistency: All references updated across CLAUDE.md, README.md, and docs/
- Renamed agents to represent person actors for better clarity:
elixir-phoenix.md→elixir-developer.mdsemantic-data.md→semantic-data-specialist.mdux-ui-design.md→ux-designer.mdweb-frontend.md→web-developer.md- Also previously renamed:
architecture-solutions.md→solutions-architect.md,data-architecture.md→data-architect.md,documentation-technical.md→technical-writer.md
- Updated all agent descriptions to use "Use PROACTIVELY for:" instead of "Use for:" for better automatic delegation with Claude Code
- Updated all documentation to reflect new agent names
- Updated agent frontmatter format: replaced
model_recommendationwithmodelfield for cleaner syntax - Agents that recommend Opus model now use
model: opusinstead of verbose recommendation text - Affected agents: brainstormer, architecture-solutions, database-specialist, performance-optimizer, security-specialist, test-architect
- Added 6 new specialized agents
- Standardized documentation across all agents
- Improved agent collaboration patterns
- MIT License
- Initial version updates
- Initial release of Prometheus Agents
- 17 specialized Claude SubAgents
- Core documentation and structure