Harness Kit skill
Harness Kit is a toolkit for working with AI agents: repository audit, TDD, root-cause analysis and the odin orchestration CLI.
- 107 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The orchestrator launches agents that carry out tasks and can change repository files.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/harness-kitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The orchestrator launches agents that carry out tasks and can change repository files.
This is third-party code. Review the repository files before installing.
What it does
Harness Kit combines three parts. The first is "Pattern Engineering": skills and subagents in `.claude/` (`hk-arch-audit`, `hk-autonomy-audit`, `hk-changelog`, `hk-merge-resolve`, `hk-local-diagnose` and others) that introduce TDD-first execution, structured root-cause analysis, knowledge compounding and cost-aware delegation. The other two parts are `odin`, a CLI that orchestrates several agents (Claude, Codex, GLM, MiniMax and more) as a dependency graph, and `taskit`, a task board with kanban, DAG view, timeline and cost analytics. The quickstart needs no install: the agent is given `docs/Quickstart.md`, audits eleven areas, produces a scored report and waits for approval of each adoption.
Who it is for. Team leads and developers building a workflow around several agents who want their repository graded against a ready checklist.
Good fit when
- A scored report on how ready a repository is for agent work is needed.
- Work is split into tasks for different agents with dependencies between them.
- After a failure a structured root-cause pass is wanted instead of guessing.
- Knowledge from finished tasks should accumulate in the project.
Not a fit when
- One small skill is wanted without a task board and orchestrator.
- The project cannot tolerate experimental software.
- Agents run without an isolated environment.
Example request
Read ../harness-kit/docs/Quickstart.md, audit this repository, show the scored report and wait for my decision on each item.Limitations
The project is experimental: the authors say they use it daily but edges are rough. The full kit starts a backend, dashboard and workers with `./dev.sh`, so it needs time and free ports (the README names 9100 and 9200). Platform notes and known gaps are in `docs/guides/forkd-setup.md` and the subproject READMEs. Running several agents in parallel spends tokens noticeably faster than a single one.
How to disable. For the skills, delete the `hk-*` folders from `.claude/skills` and the subagents from `.claude/agents`. For the full kit, stop `./dev.sh` and delete the cloned directory.
FAQ
Is anything needed for the quickstart?
No. `git clone https://github.com/deepklarity/harness-kit.git ../harness-kit` and a prompt that points the agent to the Quickstart are enough.
What does the audit check?
Eleven areas, including agent entrypoints, testing discipline, knowledge compounding and verify gates. The result is a scored report.
What is odin?
An agent orchestration CLI: plan, assign, execute, reflect. Provider checks run with `odin doctor`.
What is taskit?
A task board with an API: kanban, dependency graph, timeline and cost analytics. Each task carries proof of work.
Which agents are supported?
Whichever you have: Claude, Codex, GLM, MiniMax and others.
Discussion
No questions or reviews yet.
Related
SkillsA Rust app for macOS, Linux and Windows: skills, MCP servers, plugins, hooks and configs of 17 AI agents in one window, with a security audit
Prompt kit and plugin for Codex and Claude Code that turn repeated agent mistakes into repository rules.
A plugin with 41 skills and agents: harness engineering, a spec-to-integration pipeline and three loops that enforce constraints
The /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 Codex plugin that builds a team of subagents and skills for a project: an orchestrator splits tasks and checks the result.
README in short
Three components: a set of patterns, the odin CLI and the taskit board. The quickstart runs through a repository audit with no install.