claude-harnessOurs
Use to scaffold a complete agent-native harness: the CLAUDE.md, rules, skills, hooks, sub-agents, and memory layout, so a new project starts with the full discipline stack instead of an empty repo.
- Score
- 22.5571 signal
- Evidence
- 1 star
- Last commit
- as last read from GitHub; most reads are from 2 Sep 2026 or later
- Listed
Install
No one-command install. Set it up from its source.
What it is
A CLI that scaffolds the whole agent-native harness for a project: the operating rules (CLAUDE.md / AGENTS.md), a rules directory, a skills directory, lifecycle hooks, sub-agent definitions, and the memory layout. It is the "whole harness in a box": the assembled form of every other component in this brain.
When to use it
When starting a new agent-native project and you want the proven discipline stack from day one, not an empty repo you'll bolt rules onto later. The trigger is "set up the harness for this repo."
How to install / invoke
Install globally and run it in a new project directory to lay down the harness scaffold. Then prune to what the project needs and run a harness review to keep it lean.
Notes
The harness is the product. Start from the full scaffold, then apply Simplicity-First: keep the components that
earn their place and remove the rest. Pair with harness-review and agentbench to keep it healthy and scored.
Related
karpathy-coding-discipline
Drop into CLAUDE.md/AGENTS.md as the first behavior norm a coding agent ingrains: think before coding, prefer the simplest solution, change only what you own, and execute toward the stated goal.
Not Indexed
- harness-review
- agent-file-format