Agentное проектирование с контролем качества

★ 7.8 · ai-agents

agentic-engineering is a Claude Code skill that structures engineering workflows where AI agents handle most implementation work while humans enforce quality gates and risk controls. The skill is built on four operating principles: eval-first execution (define completion criteria before running anything), decomposing work into independently verifiable 15-minute units each with a single dominant risk, routing tasks to Haiku, Sonnet, or Opus based on complexity, and running regression evals after every implementation step. It includes a code-review checklist for AI-generated code covering edge cases, error boundaries, security assumptions, and hidden coupling, plus a per-task cost tracking template for model tier, token estimates, retries, and wall-clock time. Teams looking to run AI-driven development with predictable quality and controlled token spend will find this skill directly applicable.