AI engineering — MCP servers · agent harnesses · context engineering

  Built, not just used — a production MCP server (20 tool modules, typed contracts, official TS SDK). Custom Claude Code hooks, skills, git-backed agent memory.
  Agent harnesses — multi-agent orchestration, parallel fan-out, adversarial verification, worktree isolation.
  Context engineering — progressive disclosure over flat prompts; token budgets as an engineering constraint.
  Spec-driven delivery — AI code review, test generation and OpenAPI contracts inside a real CI pipeline.

_{Claude Code · Codex CLI · Cursor · Copilot · MCP · Anthropic / OpenAI / Gemini APIs · RAG · LLM evals}

_{Most of the work lives in private and corporate repositories — the graph reflects volume, not the code.}