Harness Starter Kit plugin
Prompt kit and plugin for Codex and Claude Code that turn repeated agent mistakes into repository rules.
- 116 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- By prompt the agent creates and edits repository files, including `AGENTS.md` and the `.harness` folder.
- The `review sub-agent` command may start a separate read-only subagent when the runtime allows it.
Install
- Works in
Manual install
codex plugin marketplace add harnessworks/harness-agent-skills-marketplace --ref v0.1.16The first install command from the README. Then restart Codex and install `harness-agent-skills` on the Plugins screen.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- By prompt the agent creates and edits repository files, including `AGENTS.md` and the `.harness` folder.
- The `review sub-agent` command may start a separate read-only subagent when the runtime allows it.
This is third-party code. Review the repository files before installing.
What it does
Harness Starter Kit turns typical coding-agent mistakes into durable repository material: instructions in `AGENTS.md`, checks, decision and failure records, evaluation tasks. The main path is prompt-based. The agent reads the kit, inspects your repository and adds only the minimum needed parts without overwriting existing files. The `harness-agent-skills` plugin for Codex and Claude Code bundles six skills under one `harness` command. `doctor` scores repository readiness without edits; `adopt` applies a minimal set; `review` challenges the current diff before a commit. `update` pulls in a newer kit version, and `refresh` hunts stale and duplicated rules.
Who it is for. Teams whose agents keep editing the wrong files or breaking project conventions, and repository owners who want checkable rules instead of one long prompt.
Good fit when
- An agent repeats the same mistake and it is time to pin it with a rule or a check.
- A repository has no `AGENTS.md` and needs a readiness check before the first agent run.
- Before a commit you want an independent review of the diff from the agent-discipline angle.
- Stale and duplicated rules piled up and must be cleaned with approval per file.
- A new kit version is out and has to be pulled in selectively.
Not a fit when
- You want a ready setup in one run: the kit is not an automatic installer.
- You need a proven quality gain for the agent, which the kit does not promise and asks you to measure separately.
- The repository is throwaway and agent rules will not be reused.
Example request
/harness doctor: assess how ready this repository is for agent work and change nothing.Limitations
The kit is prompt-based: behaviour depends on the model and on how carefully the agent reads the repository. The `/harness ...` commands in Cursor chat and other editors work as request text and appear in the command palette only after manual setup. `doctor` checks the health of instructions but does not prove that agent mistakes decrease: that is measured separately through task outcomes. The agent may edit `AGENTS.md`, so review changes in the diff before committing.
How to disable. For the plugin run `claude plugin uninstall harness-agent-skills` or disable it on the Codex plugins screen. Remove `AGENTS.md`, `.harness` and the `harness-starter-kit` folder from your repository by hand.
FAQ
How do I install Harness Starter Kit in Claude Code?
Run `claude plugin marketplace add harnessworks/harness-agent-skills-marketplace@v0.1.16`, then `claude plugin install harness-agent-skills@harnessworks`. The `/harness-agent-skills:harness` command becomes available.
How do I install the kit in Codex?
The command `codex plugin marketplace add harnessworks/harness-agent-skills-marketplace --ref v0.1.16` adds the marketplace. Restart Codex, open the Plugins screen, select `Harnessworks` and install `harness-agent-skills`.
Can I skip the plugin?
Yes. Open the repository in an agent and paste the prompt from the Quick Start section. The agent clones the kit into `./harness-starter-kit` and applies the workflow itself.
What does the doctor command do?
It checks durable signs of order in the repository and prints a report without changing anything. Then `adopt` makes minimal edits, and `review` examines the current diff before a commit.
Does refresh delete old rules?
No, it deletes, archives or renames nothing without explicit permission for each file. The command only finds stale and duplicated places.
Discussion
No questions or reviews yet.
Related
PluginsThe /better-harness command reviews how an AI agent's work is set up in a project: goals, checks, delivery and memory, and returns a prioritized fix report
One hundred ready agent teams for Claude Code, each with specialists, an orchestrator skill and domain skills, in English and Korean
A meta-skill for Claude Code that turns a domain description into an agent team and the skills it uses, using six team-architecture patterns
The Harness MCP server: one agent starts other Claude Code and Codex sessions on your computer, waits for them, reads their answers and closes them
One config for Claude Code and Codex in every repo: shared rules, skill packs, second-model review and a destructive-command guard
A plugin and kit for Claude Code and Codex: agent working principles, safety hooks, skills and CLAUDE.md templates in one repository
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Harness Starter Kit (this card) | no | 116 | |
| Repository Harness | no | 1,264 | puts AGENTS.md, a document map and onboarding skills into the repository; Harness Starter Kit adds readiness scoring, failure records, checks and a benchmark |
| Better Harness | no | 2,443 | reviews how agent work is organised with one command and returns a report; Harness Starter Kit runs a five-command process from scoring to updates |
README in short
The repository combines harness-engineering theory, a prompt for applying the kit to another repository, profile templates for popular stacks and benchmark tasks. A separate `agent-skills` folder holds the plugin for Codex and Claude Code.