inferred from GitHub

LABS

Lazy-loading ECC fork — full Claude Code enhancement suite with on-demand component loading via registry-driven auto-detection

Code Cerebral Valley

Core trick

Loads a compact trigger-table registry at session start and fetches each skill or agent file only when a request matches.

Evidence

README matched ""Cerebral Valley" "Opus 4.6"": docs/ko-KR/README.md

README excerpt

# LABS

**Lazy-loaded Agent/skill Base System for Claude Code.**

LABS is a performance fork of [everything-claude-code](https://github.com/affaan-m/everything-claude-code) that replaces eager component loading with a dual-format registry architecture. Instead of dumping every skill, agent, and command into context at session start, LABS loads only a lightweight registry (~800-1200 tokens) and resolves components on demand.

## Install

```
claude plugin add LuciferDono/LABS
```

## How It Works

At session start, Claude reads only `REGISTRY.md` -- a compact trigger-table covering all 259 components. When a user request matches a trigger, the corresponding skill/agent/command file is loaded just-in-time from the path in `registry.json`.

![Architecture](ecc_lazy_loader_architecture.svg)

**Two files, one system:**

- `REGISTRY.md` -- human-readable trigger table, loaded into context at session start
- `registry.json` -- machine-readable manifest with file paths, used by the loader at resolve time

## Components

| Type | Count |
|------|-------|
| Skills | 146 |
| Agents | 38 |
| Commands | 72 |
| Contexts | 3 |
| **Total** | **259** |

## Key Commands

| Command | Purpose |
|---------|---------|
| `/plan` | Plan before implementing |
| `/tdd` | Test-driven development workflow |
| `/verify` | Run all quality gates |
| `/code-review` | Review code before committing |
| `/build-fix` | Fix build and type errors incrementally |
| `/orchestrate` | Delegate complex tasks across ag

Core trick

Evidence

Tags

README excerpt