Agentic Engineering Toolkit
Custom MCP servers, prompt architectures, and structured skill systems enabling Claude Code agents to operate autonomously across multiple codebases.
Overview
A suite of agentic engineering tools built to accelerate real-world software development. The toolkit includes MCP (Model Context Protocol) servers in both Python and Java that serve multi-codebase domain knowledge via RAG with pgvector embeddings, enabling AI agents to reason about business logic across project boundaries.
The system uses structured skill and context file architectures (CLAUDE.md, SKILL.md, AGENT.md) that define boundaries, conventions, and capabilities for autonomous coding agents. These aren't theoretical — they're battle-tested on production codebases where they measurably reduced development time and review cycles.
Tech Stack
Deep Dive
Key components include prompt architectures for feature implementation, code review, and workflow automation; a cognitive space system for organizing agent memory by project and domain; and delegation patterns that allow a human operator to task multiple agents working in parallel across isolated project workspaces.
Built an internal AI support agent for production triage that could query logs, reference historical incidents, and suggest diagnostic steps.
Key Outcomes
- MCP servers in Python and Java serving multi-codebase domain knowledge via RAG
- Structured skill/context file systems for autonomous agent operation within project boundaries
- Prompt architectures for feature implementation, code review, and workflow automation
- Reduced non-trivial feature development from ~4 weeks to ~2 weeks on production codebases
- Internal AI support agent for production triage and diagnostics
- Multi-agent delegation patterns for parallel work across isolated project workspaces