Add nv:context – Claude Code skill for context engineering

[johnnichev]

Summary

nv:context is a Claude Code skill that automates the setup phase for AI coding agents—turning a manual, error-prone workflow into a 3-minute interview that generates production-ready configs.

What It Does

• Interview-driven setup: Asks about your tools, codebase constraints, and pain points
• Parallel code auditing: Subagents scan your repo for non-obvious patterns, landmines, and context pitfalls
• Hierarchy of Leverage scoring: Evaluates 6 layers (repository structure, tool integration, memory artifacts, hook strategy, session management, persona/tone) and scores current state
• Generates only what matters: CLAUDE.md, AGENTS.md, hooks, or session management configs—no unnecessary bloat

Why It Works

Context engineering isn't just "put the right info in the prompt." It's architecture. nv:context applies the Hierarchy of Leverage framework systematically:

1. Tool detection (auto-discover what's actually available)
2. Constraint discovery (find what breaks agents silently)
3. Memory efficiency (pack context windows optimally)
4. Recovery patterns (hooks and session state for reliability)
5. Persona alignment (tone, decision-making style, guardrails)

Production Results

Deployed on 3 real codebases with measurable impact:

• selectools: CLAUDE.md -85% (440→67 lines), leverage +9 points (49→58/60)
• nichevlabs: SESSION.md -93% (805→59 lines), saved 15.8K tokens per session
• sheriff: leverage +6 points (36→42/60) through iterative refinement

Installation

npx skills add johnnichev/nv-context -g -y

Run /nv-context in any Claude Code session.

📚 Full docs: https://skills.nichevlabs.com
💻 Source: https://github.com/johnnichev/nv-context