HAR CLI
Gives each coding agent its own repository copy with separate ports and a database, then verifies results in one shared pipeline.
- 129 ★ on the repository
- Last commit
Broad access
Why this level
- Runs project commands, dev servers and database containers inside separate working copies.
- Creates branches and writes the `.har/` directory into the repository.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/harDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Runs project commands, dev servers and database containers inside separate working copies.
- Creates branches and writes the `.har/` directory into the repository.
This is third-party code. Review the repository files before installing.
What it does
HAR is a CLI and MCP server that turns a repository into a set of isolated workspaces for agents. Each workspace is a fresh copy of the project on its own branch, with its own ports and, when the project needs one, its own database. The main checkout stays untouched. Launch, verify and teardown stages live in a `.har/` directory inside the repository. Claude Code, Cursor and Codex read that contract the same way instead of piecing rules together from README files, CLAUDE.md and CI configs. Verification runs the project's own checks and keeps logs, artifacts and a hash of the validated tree, so a reviewer sees proof rather than the agent's self-report. The package also ships Mission Control, a local dashboard for all repositories, workspaces and runs, and plugins with ready verification bundles such as Playwright and RocketSim.
Who it is for. Teams running several coding agents on one repository who are tired of clashing ports, databases and branches.
Good fit when
- Two or three agents edit one project and a shared dev server with a shared database breaks their runs.
- A reviewer wants to see which checks passed and on which code instead of trusting the agent's summary.
- Launch rules are scattered across README files, Cursor rules and CI, and agents read them differently.
- Launch scripts drifted after a new dependency and the drift should surface before a run fails.
Not a fit when
- There is one agent, the project is small and a plain git branch is enough.
- The repository cannot run locally without external services that cannot be stood up per copy.
- Kernel-level isolation for untrusted code is needed: HAR separates working copies and does not replace a sandbox.
Example request
Launch a separate workspace for the payments task, make the change, run the full verification and show what passed.Limitations
The project must be prepared first: `har onboard` creates the `.har/` directory, and launch and verify commands need adapting to your stack. The package targets Node.js 20 and newer. Per-workspace databases and services start through Docker Compose when the project is set up that way, so heavy stacks consume a lot of memory when several workspaces run at once. Verification is only as honest as the project's tests: HAR runs them and records the result but does not invent new ones. Installation runs the `ensure-telemetry-default.cjs` postinstall script, so read it in the repository before installing. The project moves fast, so check commands against the documentation at harproject.dev.
How to disable. Remove the `har` entry from the client's MCP configuration, run `npm uninstall -g @osfactory/har` and delete the `.har/` directory from the repository. Close workspaces and branches with HAR commands before uninstalling.
FAQ
How do I connect HAR to Cursor or Claude Code?
Register the MCP server with `har mcp --repo <path to repository>`. The repository includes a `.cursor/mcp.json.example` for Cursor. After that the agent asks HAR which scripts and checks the project has.
What does har onboard create?
It prepares the `.har/` directory with the project contract and prints a prompt for the coding agent to finish the setup.
What does a full verification mean?
`har env verify 1 --full` runs the project's real checks and also records the state of the whole codebase, so a pass is tied to a specific tree. HAR enforces that result at commit time as well.
How do I notice that scripts went stale?
`har env maintain` compares the installed harness with current templates and reports the drift before it causes a silent failure.
Which plugins does HAR have?
They are verification bundles for specific tools, for example Playwright and RocketSim: a plugin registers its stages in the project's harness.
Discussion
No questions or reviews yet.
Related
CLI toolsA CLI, hooks and skills for long Claude, Codex and Pi work: the issue and PR own the task and review gets a PR with checks
Agent Harness Kit: structured agent workflow
Agent Harness Kit
The ahk CLI sets up five agent roles, a task MCP server and spec skills in your project for Claude Code, Codex and Cursor
Repository Harness installs an AGENTS.md entrypoint, a document map and explicitly invoked onboarding skills into a repo and updates them safely
Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook
A CLI that generates and maintains AI context files, CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions, keeping them accurate as code changes
One desktop window for Claude Code, Codex, Gemini CLI and OpenCode: terminals, pull requests, diffs and approvals
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| HAR (this card) | no | 129 | ||
| container-use | yes | not required | 4,087 | gives an agent a container and a git branch; HAR adds the `.har/` contract, hash-bound verification and the Mission Control dashboard |
| Claude Squad | no | 8,738 | manages agent sessions in tmux and worktrees; HAR focuses on verifying results and reproducible stack launch |
README in short
The author calls trusting an agent's output without manual re-checks the hard part. HAR addresses five gaps: no standard way to run and verify a repository, agents colliding, results needing re-verification, a contract tied to one platform, and hand-rolled scripts that rot. Work follows five steps: discover, isolate, build, verify, hand off a branch with evidence.