agentic-engineering

The agentic-engineering skill enables AI to operate as a primary executor in engineering workflows, allowing human operators to focus on quality assurance and risk management. By leveraging an eval-first approach, task decomposition, and cost-aware model routing, this skill ensures that complex engineering tasks are broken down into manageable units, evaluated for correctness, and executed efficiently across different model tiers. It addresses the challenge of scaling AI-driven implementation while maintaining control over critical software quality aspects, such as correctness, security, and reliability.

Key capabilities include structured task decomposition into independently verifiable units, application of model routing based on task complexity, and continuous evaluation loops to monitor performance and regression. The skill supports session management strategies to maintain continuity across closely-related tasks or reset for major phase transitions. Additionally, it emphasizes review priorities, focusing on invariants, edge cases, error boundaries, and potential security risks while avoiding wasted effort on style-only issues. Cost discipline is integrated, tracking metrics such as model choice, token usage, retries, wall-clock time, and task success or failure, ensuring efficient use of computational resources.

This skill is ideal for AI engineers, technical leads, and development teams who want to offload routine implementation work to AI agents while retaining oversight of quality-critical tasks. It is particularly useful in scenarios involving multi-file projects, complex architecture analysis, and code refactoring where human verification is essential. By providing structured guidelines for evaluation, decomposition, and model tier escalation, agentic-engineering helps teams implement robust, high-quality software faster and with more predictable outcomes.