proagent-agentic-engineering

Create Claude Code workflows using the Workflow Factory -- complexity-assessed, template-driven, contract-validated artifact generation

ProAgent Agentic Engineering hub -- show capabilities, route to sub-commands for building agent systems

Review agent systems, skills, prompts, commands, hooks, plugins, orchestration patterns, and MCP servers for quality and best practices

Execute agentic engineering workflows -- create agents, skills, commands, hooks, MCP servers, plugins, orchestrate multi-agent systems, and evaluate agents

Expert in designing, building, orchestrating, and optimizing AI agent systems using Claude Code. Specializes in creating sub-agents, skills, slash commands, hooks, MCP servers, and plugins following Anthropic's official specifications. Deep knowledge of multi-agent orchestration patterns (sequential pipeline, parallel fan-out, hierarchical delegation, state machine), context engineering (3-file planning, context priming, knowledge graphs), plugin lifecycle management, and thinking model commands. Routing knowledge for autonomous coding loop frameworks (ralph-orchestrator, casdk-harness, Auto-Claude, gastown -- see references/external-frameworks.md for confidence levels), multi-backend adapters (Claude, Kiro, Gemini, Codex -- routing-level, see references/external-frameworks.md), spec-driven development (AWOS, ProAgent SDLC pipeline -- routing-level), and Trust Ladder progressive autonomy (ZTE 5-level -- routing-level, see references/external-frameworks.md). Use PROACTIVELY when user requests: - Creating or reviewing agents, skills, commands, hooks, plugins, or MCP servers - Designing multi-agent orchestration workflows - Prompt engineering or context engineering for agent systems - Evaluating agent effectiveness or quality - Building Claude Code extensions or customizations - Setting up autonomous coding loops (ralph, casdk-harness, Auto-Claude, gastown) - Configuring multi-backend LLM adapters - Spec-driven development workflows (AWOS, ProAgent SDLC) - Plugin lifecycle management and tool registries - Applying thinking models and structured reasoning to agent design Examples: <example> Context: User wants to create a specialized agent for their domain user: "Help me create a Terraform infrastructure agent that can plan and apply changes safely" assistant: "I'll use the agentic-engineering-specialist to design a comprehensive Terraform agent with proper tool restrictions, safety constraints, and discovery-optimized description." <commentary> Creating domain-specific agents requires understanding of agent architecture, tool selection, and discovery optimization -- core expertise of this specialist. </commentary> </example> <example> Context: User needs to orchestrate multiple agents for a complex workflow user: "I need to set up a code review pipeline with security, performance, and style reviewers running in parallel" assistant: "I'll use the agentic-engineering-specialist to design a parallel fan-out orchestration pattern with specialized reviewer agents and an aggregator." <commentary> Multi-agent orchestration requires selecting the right pattern, defining hand-offs, and managing token budgets across agents -- this specialist handles the complete design. </commentary> </example> <example> Context: User wants to add automation hooks to their Claude Code setup user: "How do I automatically format Python files after Claude edits them?" assistant: "I'll use the agentic-engineering-specialist to create a PostToolUse hook with a matcher for Edit|Write and a shell command that runs Black on modified Python files." <commentary> Hook configuration requires understanding lifecycle events, matcher patterns, stdin JSON processing with jq, and security implications -- this specialist knows the patterns. </commentary> </example> <example> Context: User needs to build an MCP server user: "I want to create an MCP server that connects Claude to our internal Jira API" assistant: "I'll use the agentic-engineering-specialist to guide you through the four-phase MCP server development process: research, implementation, review, and evaluation." <commentary> MCP server development requires agent-centric design principles (workflow-oriented tools, context optimization, actionable errors) and evaluation-driven development -- core expertise of this specialist. </commentary> </example> <example> Context: User wants to set up an autonomous coding loop user: "I need to configure an autonomous agent loop that can implement features end-to-end without manual intervention" assistant: "I'll use the agentic-engineering-specialist to help choose between ralph-orchestrator, casdk-harness, Auto-Claude, or gastown, configure the appropriate orchestration preset, set trust ladder levels, and define exit conditions." <commentary> Autonomous coding loop setup requires selecting the right framework, trust/autonomy configuration, context persistence strategies, and confirmation gate design -- this specialist routes to the right framework (routing-level knowledge -- see references/external-frameworks.md for details). </commentary> </example>

Comprehensive skill for designing, building, orchestrating, and optimizing AI agent systems with Claude Code. Covers agent design patterns, multi-agent orchestration, skill/command/hook creation, MCP server development, prompt engineering, plugin architecture, agent evaluation, context engineering, autonomous coding loops, multi-backend orchestration, trust/autonomy frameworks, spec-driven development, thinking model commands, and the Workflow Factory for template-driven artifact creation with complexity assessment and inter-artifact contracts. This skill should be used when the user mentions: - Creating agents, skills, commands, hooks, or plugins for Claude Code - Multi-agent orchestration, parallel agents, subagent workflows - MCP server development or integration - Prompt engineering, context engineering, or prompt-driven development - Agent evaluation, testing agent systems, or agent quality assurance - Claude Code extensions, customization, or configuration - Agentic development workflows (ADW), trust ladder, progressive autonomy - Subagent-driven development (SDD), dispatching parallel agents - File-based planning with hooks, plan execution with agents - Autonomous coding loops (ralph, casdk-harness, Auto-Claude, gastown) - Multi-backend adapters (Claude, Kiro, Gemini, Codex) - Spec-driven development (AWOS spec/implement/verify, ProAgent SDLC pipeline) - Tool registry, plugin lifecycle, knowledge graph integration - Thinking models, first-principles reasoning, mental model commands - 3-file planning pattern, context handoff, session persistence - Workflow factory, create workflow, generate artifacts - Complexity assessment (simple, medium, complex) - Design decisions, mandatory design gates - Expert system trio (plan, build, improve), self-improving workflows - Inter-artifact contracts, producer-consumer data passing - Report types (summary, diff, handoff, progress, audit, path-only) - Error handling patterns (gate guard, step fallback, graceful degradation) - Naming conventions, artifact templates, coherence validation Do NOT use for: - General software development without an agentic component (use sdlc or backend/frontend skills) - DevOps infrastructure without agent involvement (use devops skills) - Pure testing without agent evaluation (use qa skills)

3 hooks across 3 events