Agent & Harness Development

Create new skills, improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, or benchmark skill performance with variance analysis.

Create and evaluate new skills

Plugin development toolkit with skills for creating agents, commands, hooks, MCP integrations, and comprehensive plugin structure guidance

Plugin development toolkit

Automatically converts any documentation website into a Claude AI skill in minutes

Find and reuse skills from your environment

Authors new skills for this library and grades existing ones against the repo's eval template. Operates in four modes: Create, Grade, Compare, Analyze.

Skill authoring agent

Execute a phased implementation plan using subagents. Use when asked to execute, run, or carry out a plan — especially one created by make-plan.

Execute-then-codify task loop

Skills for designing and building MCP servers that work seamlessly with Claude — guides you through deployment models (remote HTTP, MCPB, local), tool design patterns, auth, and interactive MCP apps.

Design MCP servers that work with Claude

Connect Claude to a private MCP server through an Anthropic MCP tunnel. Drives the Docker Compose quickstart end to end: certificates, proxy config, cloudflared, and a verifiable sample server.

Expose private MCP servers via tunnels

Expert assistant for developing Model Context Protocol (MCP) servers in Python

Python MCP server patterns

Expert assistant for developing Model Context Protocol (MCP) servers in TypeScript

TypeScript MCP server patterns

A comprehensive example plugin demonstrating all Claude Code extension options including commands, agents, skills, hooks, and MCP servers

Reference plugin structure

Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review

End-to-end feature development workflow

Agent Designer - Multi-Agent System Architecture

Design agents with clarity and scope

Design and optimize AI agent action spaces, tool definitions, and observation formatting for higher completion rates.

Build agent harnesses

Claude Agent SDK Development Plugin

Claude Agent SDK development

Continuous self-referential AI loops for interactive iterative development, implementing the Ralph Wiggum technique. Run Claude in a while-true loop with the same prompt until task completion.

Iterative agent loops

Easily create hooks to prevent unwanted behaviors by analyzing conversation patterns

Create hooks from conversation patterns

Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.

Audit and improve CLAUDE.md files

Analyze codebases and recommend tailored Claude Code automations such as hooks, skills, MCP servers, and subagents.

Tailored Claude Code setup for a repo

Seamless onboarding for the Code-with-Claude Makers Cardputer: one /maker-setup command clones the build-with-claude repo, flashes UIFlow firmware, and installs the Claude Buddy app bundle onto a freshly-plugged-in M5Stack Cardputer-Adv.

Makers Cardputer onboarding

This skill should be used when the user asks to "implement LLM-as-judge", "compare model outputs", "create evaluation rubrics", "mitigate evaluation bias", or mentions direct scoring, pairwise comparison, position bias, evaluation pipelines, or automated quality assessment.

Rigorous evaluation of agent output

Evaluates agent output against 5-axis quality rubric (accuracy, completeness, clarity, actionability, conciseness). Use after any non-trivial task when the user wants a quality assessment, or when the agent-self-evaluation skill is active. Produces structured scorecard with evidence and improvement suggestions.

Judge agent performance

Performance-obsessed, benchmark-driven analysis demanding measured evidence (Matteo Collina inspiration)

Multi-perspective benchmarking

Solve competition math (IMO, Putnam, USAMO) with adversarial verification that catches what self-verification misses. Fresh-context verifiers attack proofs with specific failure patterns. Calibrated abstention over bluffing.

Adversarial verification for hard problems