Overview

Harness is an architectural tool that utilizes Claude Code’s agent team system to break down and coordinate complex tasks into specialized agent teams. When you say “Configure the harness,” it automatically creates agent definitions (.claude/agents/) and skills (.claude/skills/) that fit your domain.

Claude et al

Differences between original harness vs harness-init (Cursor environment)

📁 Differences in settings file recognition path

• Claude Code: Reads and interprets .claude/agents/, .claude/skills/, CLAUDE.md rules directly at the system level.
• Cursor: By default, it uses the .cursorrules file or the .cursor/ directory as the context system.

🛠️ Execution loop and tooling constraints

• The essence of the harness framework is not simply creating markdown documents, but a terminal-based execution loop that issues terminal commands, self-corrects if an error occurs (Self-Correction), and rolls back with a git command.
• Claude Code runs an isolated pipeline with completely autonomous shell command authority in the CLI environment, but Cursor’s Agent (Composer, etc.) moves based on the IDE context, so the powerful system control or automatic ratchet mechanism required by the harness may be interrupted or behave differently.

Cursor Pre-Setup

project_home="$(pwd)"
workspace_home="$HOME/workspace"
cd $workspace_home
git clone https://github.com/Gizele1/harness-init

cd $project_home
mkdir -p .cursor/rules/harness-init
cd .cursor/rules/harness-init
cp -r "$workspace_home/harness-init/skills/harness-init/SKILL.md" .
cp -r "$workspace_home/harness-init/skills/harness-init/references" .

Common Post-Setup

text to

• “harness init this repo”
• “make this repo agent-ready”
• “set up architecture boundaries”