Agentic Engineers — AI skill for Claude Code

Agentic Engineers

A **multi-agent orchestration framework** for coordinating specialized AI agents through structured handoffs, quality gates, and cost-aware model selection. Designed for integration with coding CLIs: **OpenCode**, **Copilot**, **Claude**, **Codex**.

What It Is

Agentic Engineers solves the multi-agent coordination problem: how do you route work to the right specialist, enforce quality consistently, and keep cost proportional to task complexity — through an elegant, minimal orchestration layer?

**The answer:** an ORCHESTRATOR-FIRST architecture built on **direct sub-agent spawning**:

1. Work is expressed as a DELEGATE task (structured YAML: scope, context, plan, success criteria)
2. The Orchestrator spawns the right specialist directly (Agent/Task tool) and reads the HANDBACK back as that spawn call's result synchronously, in-context
3. The specialist executes and returns a HANDBACK with results + metrics
4. The harness session transcript itself — every DELEGATE as a spawn prompt, every HANDBACK as that spawn's result — is the durable audit record
5. Metrics feed back into model selection and routing for future tasks

Goals

The framework is built to be **minimal, portable, and self-reducing**:

• Small AGENTS + SKILLS mechanisms — a portable orchestration layer that works across harnesses (Claude, Copilot, OpenCode, Codex) without proprietary integrations
• Framework self-reduction — as base and frontier models improve, LOC and complexity are meant to decrease, not grow
• Eventual redundancy — when harnesses compose and delegate work well by default, this coordination layer should fade into standard practice
• Harness comparative analysis — the src/ → make render → dist/ → make install → ~/. pipeline lets the same agent/skill roster be compared across harnesses for feature parity and quality trade-offs

Positioning

Agentic-engineers is one layer in a three-tier orchestration landscape: **heavy frameworks**