skills.rest
khetansarvesh
khetansarveshCOMMUNITY
Explore all skills by this author →
GitHub

agentic-engineering

Execute eval-first engineering workflows with cost-aware model routing.

1|Updated Mar 6, 2026
Search Tags:#task-decomposition#regression-testing#model-routing#agentic-engineering#secure-code-review#eval-first#cost-discipline
One-click install
npx skills add https://github.com/khetansarvesh/ai_skills_repo --skill agentic-engineering-khetansarvesh
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agentic-engineering
Source: https://github.com/khetansarvesh/ai_skills_repo/tree/main/skills/agentic-engineering
Command: npx skills add https://github.com/khetansarvesh/ai_skills_repo --skill agentic-engineering-khetansarvesh

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It helps teams complete complex engineering tasks faster and safer by combining agentic implementation with predefined quality gates, eval-driven regression checks, and cost-aware model routing.

Core Features & Use Cases

  • Eval-first execution loop: Define capability and regression evals, establish failure signatures, implement, then re-run evals to compare deltas.
  • Risk-focused decomposition: Break work into independently verifiable units with clear done conditions using the 15-minute unit rule.
  • Cost-aware model routing: Use smaller models for narrow edits and classification, escalate to stronger models for refactors, and reserve the top tier for architecture and root-cause analysis.
  • Quality and safety review focus: Emphasize invariants, edge cases, error boundaries, security/auth assumptions, and hidden coupling over style-only disagreements.
  • Operational session strategy: Continue sessions for tightly coupled steps, restart after major phases, and compact after milestones rather than mid-debugging.

Quick Start

Tell your AI agent: “Plan an agentic engineering workflow for my feature, decompose it into 15-minute verifiable units, route tasks by model tier, and run evals/regressions before and after implementation.”

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 manage AI implementation tasks?▼

Eval-first agentic engineering is an execution strategy that defines capability and regression evals, establishes failure signatures, implements features, and re-runs evals to compare deltas for safe AI-driven development.

How do I decompose complex development tasks for an AI agent to execute safely?▼

You decompose complex tasks using the 15-minute unit rule, breaking work into independently verifiable units with predefined completion conditions to ensure tightly coupled steps are verifiable and safe.

Can I route AI coding tasks to different model tiers based on complexity to control costs?▼

Yes, cost-aware model routing uses smaller models for narrow edits and classification, escalates to stronger models for refactors, and reserves the top tier for architecture and root-cause analysis.

What is the best way to run regression checks when using AI for feature implementation?▼

The best way is capturing baseline failure signatures before implementation, then re-running capability and regression evals afterward to compare deltas and ensure no regressions are introduced.

When should I restart or compact an AI agent session during a multi-step engineering workflow?▼

You should continue sessions for tightly coupled steps, restart after major phases, and compact contexts after milestones to avoid losing reasoning state during active debugging.

How does secure code review work with eval-driven AI development workflows?▼

Secure code review in eval-driven workflows emphasizes checking invariants, edge cases, error boundaries, security/auth assumptions, and hidden coupling rather than focusing solely on style disagreements.

Related Skills

Explore More Software Engineering Skills→
rivaldiekaptrrrrivaldiekaptrrr

eval-harness

Define and run eval suites to validate AI task performance.

Community
Advanced
idiaz01idiaz01
3

eval-harness

Run repeatable capability and regression evaluations for Claude Code sessions.

Community
Advanced
ryasrkryasrk

agentic-engineering

Decompose software engineering tasks into verifiable units with eval-first execution.

Community
Advanced
richardnpaulrichardnpaul

agentic-engineering

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

Community
Advanced