Claude Agents Coordination — Development skill for Claude Code

Claude Code Agents Coordination

**Production-grade multi-agent coordination system with tiered delegation, institutional memory, and explicit tech-debt tracking.**

This repository provides reusable configuration layers for Claude Code that enable sophisticated agent orchestration across projects—from solo development to enterprise-scale workflows.

What This Solves

Three critical problems in Claude Code agent usage:

1. Context Amnesia: Agents forget prior work within 15-20 minutes
2. Coordination Chaos: Multiple agents overwrite each other's work
3. Delegation Limits: Manual oversight required for every decision

**Solution**: A tiered delegation architecture with persistent memory (registry + tech-debt tracking) that extends productive context windows to 2+ hours while enabling the Main Agent to consistently apply documented engineering criteria.

---

Core Architecture

Tiered Agent System

Tier 1: Workflow Orchestrators
├── code-quality      → Multi-level review chain (L1→L2→L3→L4)
├── test-engineer     → Test execution with failure triage
├── architect         → System design + ADRs + RFC lifecycle
└── ml-engineer       → End-to-end ML workflows

Tier 2: Specialized Execution
├── security-engineer → Threat modeling, OWASP scans
├── sre               → SLOs, postmortems, capacity planning
├── rfc               → Design document workflow
├── data-engineer     → Data pipelines + quality validation
├── frontend/backend  → Implementation work
├── devops            → CI/CD, infrastructure, git workflows
└── docs              → Technical documentation

Tier 3: On-Demand Specialists
├── lrl-nlp-expert        → Low-resource language NLP
├── data-viz-specialist   → Data storytelling + dashboards
└── ux-designer           → UX/design review + writing

Dual-Registry Model

**`_registry.md`** — What was done

• Tracks completed work, deliverables, and outcomes
• Main agent reads this for context before starting new tasks
• Updated after