← Back to Discover

harness-starter-kit

Prompt-based starter kit that guides a coding agent to add repository instructions, checks, memory and review steps to your repo, so repeated agent mistakes become lasting guidance. Has Codex and Claude Code skills.

Coding agents
Visit website ↗

What it does

Harness Starter Kit treats a repository as the lasting working environment for coding agents. Its idea is that every repeated agent mistake should be turned into something lasting, such as better instructions, an automated check, a failure or decision record, or a drift check. Adoption is prompt-first. You open your own repository in a coding agent and give it the kit's prompt. The agent inspects the repo and adds the smallest useful pieces, such as project instructions, a knowledge store, lightweight checks, local verification commands and an adoption report. Commands like /harness doctor, adopt, review, update and refresh are prompt conventions, not built-in editor commands. These workflows also ship as skills for Codex and Claude Code through a plugin marketplace, and an optional Python script can copy profile snippets into a repo before the agent adapts them. The README separates how well a repo is set up from whether agents really make fewer mistakes, and says the kit cannot prove the latter by itself, so it provides templates for measuring outcomes. The project is MIT-licensed.

Capabilities

  • A prompt that has your coding agent adapt the kit to your own repository
  • /harness doctor to check readiness without changing files
  • /harness review to challenge a change set before commit or PR
  • Templates for decision and failure memory and for effectiveness reports
  • Skills for Codex and Claude Code, plus an optional Python installer for profile snippets

When to use it

  • A coding agent keeps repeating the same mistakes in your repository
  • You want shared agent instructions and checks that live in the repo itself
  • You want a second look at an agent-made change before it is committed