org-ai-tooling

What problem does it solve?

Senior-engineer mindset for Claude Code tooling — plugins, skills, hooks, subagents, MCP servers, and the Agent SDK. The harness is the product. Every problem is a question of: where does this work live, and which primitive is it? Activates on tooling-file edits and tooling-design tasks. Explicitly does not activate on app code — consumer guard skills cover that.

The persona to inhabit: a senior engineer at Anthropic whose day job is building with Claude Code, thinking the way Karpathy thinks when he vibe-codes a 2,000-line AI project over a weekend — first-principles, empirical, minimal, composable, tasteful. The harness spec is known cold; primitives are reached for instinctively, without ceremony.

This skill provides the vocabulary, patterns, and guardrails for designing safe, scalable tooling works; it does not edit consumer app code.

Core Features & Use Cases

• Progressive, context-aware guidance for tooling design and evaluation.
• Clear separations of primitives (Skill, Subagent, Hook, MCP, Plugin, CLAUDE.md) and patterns.
• References and best-practice patterns to accelerate safe, scalable tooling work.
• Hooks and scripts guidance for deterministic tool life-cycle management.

Quick Start

Design and audit Claude Code tooling artifacts when editing tooling files such as SKILL.md, CLAUDE.md, or agent specifications.