How does the 15-minute unit rule improve AI coding?

By breaking work into units that take approximately 15 minutes to verify, it ensures the AI remains focused, risks are isolated, and the 'done' condition is always clear.

What is an eval-first loop?

It is a process where you define capability and regression evals, run them to capture failures, execute the code change, and then re-run the evals to confirm the fix without introducing new bugs.

What is the core philosophy of the Agentic Engineering skill?

The core philosophy is that AI agents should handle the bulk of implementation work while humans focus on enforcing quality, defining completion criteria, and managing risk controls.

How does model routing save costs?

It identifies task complexity and routes simple transforms to Haiku, implementation to Sonnet, and complex architecture to Opus, ensuring you only pay for the intelligence you need.

Agentic Engineering Workflow

Name: Agentic Engineering Workflow
Author: affaan-m

byaffaan-m

•

137,500

•

生产力与工作流

Optimizes AI agent performance through eval-driven execution, task decomposition, and cost-aware model routing.

The Agentic Engineering skill transforms Claude into a high-performance autonomous developer by implementing a structured, research-first development lifecycle. It shifts the human role toward high-level quality and risk control while the AI manages implementation through an 'eval-first' loop. By breaking complex engineering tasks into 15-minute verifiable units and intelligently routing work between model tiers (Haiku, Sonnet, and Opus), this skill ensures more reliable code, minimized regressions, and significant cost savings during large-scale software development.

主要功能

01Eval-first execution loop to capture failure signatures before implementation.

02Strategic session management to maintain context and prevent debugging drift.

03Cost-aware model routing across Haiku, Sonnet, and Opus tiers based on task complexity.

04Granular task decomposition using the 15-minute unit rule for verifiable progress.

05137,500 GitHub stars

06Risk-focused review protocols prioritizing invariants, edge cases, and security assumptions.

使用场景

01Optimizing development costs by automating boilerplate and classification with lower-tier models.

02Executing high-reliability refactors where regression testing is mandatory.

03Managing complex, multi-file feature implementations through structured agentic units.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add affaan-m/everything-claude-code agentic-engineering

For use in Claude.ai and ChatGPT

主要功能

01Eval-first execution loop to capture failure signatures before implementation.

02Strategic session management to maintain context and prevent debugging drift.

03Cost-aware model routing across Haiku, Sonnet, and Opus tiers based on task complexity.

04Granular task decomposition using the 15-minute unit rule for verifiable progress.

05137,500 GitHub stars

06Risk-focused review protocols prioritizing invariants, edge cases, and security assumptions.

使用场景

01Optimizing development costs by automating boilerplate and classification with lower-tier models.

02Executing high-reliability refactors where regression testing is mandatory.

03Managing complex, multi-file feature implementations through structured agentic units.