agentic-engineering

Decompose engineering tasks into verifiable units with eval-first evaluation and tiered model routing.

Updated Apr 2, 2026

Search Tags:#decomposition #engineering #agentic #model-routing #risk-controls #eval-first #cost-discipline

Ready to run

npx skills add https://github.com/richardnpaul/everything-vscode-copilot --skill agentic-engineering-richardnpaul

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agentic-engineering
Source: https://github.com/richardnpaul/everything-vscode-copilot/tree/main/.github/skills/agentic-engineering
Command: npx skills add https://github.com/richardnpaul/everything-vscode-copilot --skill agentic-engineering-richardnpaul

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Enables AI-driven engineering workflows where agents perform implementation work under human quality and risk controls, using eval-first evaluation, decomposition into agent-sized units, and cost-aware model routing to maximize reliability and traceability.

Core Features & Use Cases

Eval-first loop: define capability and regression evals, run baselines, capture failure signatures, and compare deltas to guide improvement.
Task Decomposition: apply the 15-minute unit rule to break work into independently verifiable units with clear done criteria.
Model Routing: use tiered models (Haiku, Sonnet, Opus) aligned to task complexity to optimize outcomes and resources.
Session Strategy: maintain coherent sessions for closely-coupled work, and reset after major milestones or phase transitions.
Review Focus: emphasize invariants, edge cases, error boundaries, security, and deployment considerations in AI-generated code.

Quick Start

Define a project goal, decompose it into agent-sized units, route by complexity, and run an eval-first loop to validate progress.

Frequently Asked Questions about agentic-engineering

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

What is eval-first agentic engineering and how does it control AI code generation?▼

Eval-first agentic engineering executes AI implementation work by defining capability and regression evals, running baselines, and comparing deltas to guide improvement. It controls AI code generation by capturing failure signatures and validating progress against established baselines.

How do I decompose software engineering tasks into agent-sized units?▼

Decompose software engineering tasks by applying the 15-minute unit rule, breaking work into independently verifiable units with clear done criteria. This decomposition ensures AI agents execute structured task breakdown while maintaining traceability and human quality control.

How does cost-aware model routing work for AI engineering workflows?▼

Cost-aware model routing assigns tiered models like Haiku, Sonnet, and Opus to tasks based on complexity. This routing optimizes resource allocation and outcomes by aligning model capabilities with the specific requirements of each engineering task.

Can I maintain coherent sessions across closely-coupled AI engineering tasks?▼

Maintain coherent sessions for closely-coupled AI engineering work by preserving context across related tasks, and reset sessions after major milestones or phase transitions. This session strategy ensures structured task breakdown remains reliable and traceable.

What review focus should I apply to AI-generated code for risk controls?▼

Apply risk controls to AI-generated code by emphasizing invariants, edge cases, error boundaries, security, and deployment considerations. This review focus ensures human quality control over agentic engineering workflows and validates regression checks.

What is the best way to structure an agentic engineering workflow from project goal to validation?▼

Structure agentic engineering workflows by defining a project goal, decomposing it into agent-sized units, routing tasks by complexity to tiered models, and running an eval-first loop to validate progress. This maximizes reliability and traceability across sessions.