git404hub

what is harness-for-claude fr?

ganimjeong/harness-for-claude — explained in plain English

Analysis updated 2026-05-18

30ShellAudience · developerComplexity · 2/5Setup · easy

tl;dr

Starter template that gives a new project one AGENTS.md guide, three shell scripts, and a .claude folder of hooks and settings so Claude Code and other agents follow the same routine.

vibe map

mindmap
  root((Harness-for-claude))
    Inputs
      Empty new project
    Outputs
      AGENTS.md and CLAUDE.md symlink
      Bootstrap and check scripts
      .claude hooks and settings
    Use Cases
      Standardise agent setup
      Block bad session exits
      Share PR checklists
    Tech Stack
      Shell
      Markdown
      ClaudeCode
      AGENTS.md format

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

what do people make with this?

VIBE 1

Bootstrap a fresh project with consistent agent instructions and check scripts

VIBE 2

Get the same AGENTS.md picked up by Claude Code, Codex, Cursor, and Aider

VIBE 3

Run a pre-end hook that blocks a session if lint, types, or tests fail

VIBE 4

Match PR descriptions to the same checklist the agent followed

what's the stack?

ShellMarkdownClaudeCode

how it stacks up fr

ganimjeong/harness-for-claudefree-nodes/shadowrocketchristitustech/zed-titus
Stars303029
LanguageShellShellShell
Setup difficultyeasyeasyeasy
Complexity2/51/52/5
Audiencedevelopergeneraldeveloper

Figures from each repo's GitHub metadata at analysis time.

how do i run it?

Difficulty · easy time til it works · 5min

Windows users need developer mode and core.symlinks=true or CLAUDE.md will not symlink to AGENTS.md.

in plain english

This repository is a small starter template that someone can copy when they begin a new project they expect to work on with an AI coding agent, especially Claude Code. The idea is that every new project gets the same predictable shape: one place where the agent reads its instructions, one command to install dependencies, one command to run checks, and one command to run tests. That way the agent does not have to guess where things live, and the human collaborator can rely on the same routine across projects. The agent guide is kept in a file called AGENTS.md, which is the cross-tool format that Codex, Cursor, Aider, and similar tools also read. A second file named CLAUDE.md is a symlink that points at AGENTS.md, so Claude Code picks up the same content without any duplication. The README warns Windows users that the symlink will only work if developer mode is on and a specific git setting is enabled, otherwise CLAUDE.md will appear as a plain file with the text 'AGENTS.md' inside it. Three shell scripts sit in a 'scripts' folder. The bootstrap script tries to install or prepare dependencies when it can spot a known stack. The check script runs formatting, linting, type checks, and tests when those tools are present. The test script runs only the test suite. These are written as safe defaults for a new or near-empty repo, and the author expects each team to extend them as the project grows. The template also ships a .claude folder for Claude-specific configuration. It contains a checked-in settings.json for permissions and hooks, an example local-only settings file for personal tweaks, a commands folder for project-scoped slash commands, a hooks folder with two default hooks, and an empty agents folder for project sub-agents. The two default hooks add useful context at the start of a session, including the current branch and any uncommitted work, and run the check script before the session ends, blocking the end if those checks fail. A pull request template in the .github folder lines up PR descriptions with the same checklist the agent follows.

prompts (copy fr)

prompt 1
Copy the Harness-for-claude template into my new repo and customise AGENTS.md for a Node API project
prompt 2
Fix the CLAUDE.md symlink on Windows so Claude Code picks up the AGENTS.md content
prompt 3
Extend the bootstrap script to install Python and pre-commit when it sees a pyproject.toml
prompt 4
Add a project slash command under .claude/commands that runs the check script and reports failures

Frequently asked questions

what is harness-for-claude fr?

Starter template that gives a new project one AGENTS.md guide, three shell scripts, and a .claude folder of hooks and settings so Claude Code and other agents follow the same routine.

What language is harness-for-claude written in?

Mainly Shell. The stack also includes Shell, Markdown, ClaudeCode.

How hard is harness-for-claude to set up?

Setup difficulty is rated easy, with roughly 5min to a first successful run.

Who is harness-for-claude for?

Mainly developer.

peek the repo → explain another one

This repo across BitVibe Labs

double-check against the repo, no cap.