agentic-engineering

What problem does it solve? When AI agents perform most implementation work, teams lack a structured way to define completion criteria, decompose tasks, route models by complexity, and measure quality, leading to inconsistent output and wasted cost. ## Core Features & Use Cases - Eval-First Loop: Define capability and regression evals, capture baseline failure signatures, then re-run evals after implementation to compare deltas. - Task Decomposition: Apply the 15-minute unit rule so each unit is independently verifiable with a single dominant risk and a clear done condition. - Model Routing & Cost Discipline: Route Haiku for narrow edits, Sonnet for implementation, Opus for architecture, and track model, tokens, retries, and success per task. - Use Case: An engineering lead assigns a multi-file refactor to AI agents, defines regression evals first, splits work into verifiable units, and escalates model tier only when a lower tier shows a clear reasoning gap. ## Quick Start Use the agentic-engineering skill to plan and execute this feature with eval-first checks, decomposed tasks, and cost-aware model routing.