explaingit

popschlock/claude-harness-generator

12Audience · developerComplexity · 2/5Setup · easy

TLDR

A CLI tool that interviews you about your project and generates a complete Claude Code configuration, including agents, commands, rules, and memory files, in 10 to 20 minutes.

Mindmap

mindmap
  root((harness generator))
    What it does
      Interview workflow
      Configuration files
      Agent definitions
    Profiles
      4 base profiles
      16 domain presets
      Custom path
    Domains
      Legal and finance
      Grant writing
      Security audits
    Internal Agents
      Intake interviewer
      Architecture designer
      Validator agent
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Things people build with this

USE CASE 1

Generate a complete Claude Code project setup tailored to your domain and workflow in about 15 minutes.

USE CASE 2

Pick a pre-built domain profile for legal, finance, security, or grant writing and get a configuration with appropriate safeguards.

USE CASE 3

Validate or upgrade an existing Claude Code harness against current best practices.

Tech stack

Claude Code

Getting it running

Difficulty · easy Time to first run · 30min

Requires Claude Code CLI and either an Anthropic API key or an active Claude subscription.

In plain English

Claude Harness Generator is a tool that builds a complete Claude Code project setup for you, tailored to your specific situation, by asking you a series of questions. The setup it produces is called a harness, which is the combination of configuration files, rules, specialized AI agents, custom commands, and memory files that tell Claude Code how to behave in your project. Normally, building a good harness for Claude Code requires knowing the tool well and spending time writing and testing configuration by hand. This project handles that by running an interview: you describe your domain (such as software development, legal research, data analysis, or game development), your team size, whether you use version control, and how much independence you want Claude to have. The generator then produces all the necessary files in a directory you choose. The whole process takes about 10 to 20 minutes. For people who want to move faster, the generator includes 4 base profiles covering broad categories and 16 pre-built domain profiles for more specific fields including financial modeling, grant writing, customer support, hiring pipelines, and security audits. High-stakes domains like legal and finance come with relevant safeguards built into the generated configuration. If none of the presets fit, a custom path runs a deeper interview and builds a configuration from scratch. Beyond creating new setups, the tool includes commands to validate an existing harness for structural problems, upgrade it against current best practices with your input, and refresh the generator's own knowledge base by pulling in updated guidance. The generator itself uses five internal AI agents: one that conducts the intake interview, one that designs the architecture, one that writes the output files in five focused passes, one that validates the result before handing it over, and one that analyzes existing harnesses for improvements. The project requires the Claude Code command-line tool and either an Anthropic API key or an active Claude subscription.

Copy-paste prompts

Prompt 1
Run claude-harness-generator and set up a Claude Code harness for a Python backend project with a team of 3 and moderate AI independence.
Prompt 2
Use the financial modeling domain profile in claude-harness-generator to produce a Claude Code config with built-in safeguards for high-stakes outputs.
Prompt 3
Run the validate command in claude-harness-generator against my existing .claude/ folder and list any structural problems it finds.
Open on GitHub → Explain another repo

← popschlock on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.