BenchKeeper

BenchKeeper sets up Claude Code and Codex the same way on your machine and in each repository: short global rules and a workflow skill that sizes each task, plus a dev container, AGENTS.md and skills per repository. Its app shows every change before it applies it and records each run so you can undo it.

What BenchKeeper is: the workflow, what each part brings and the defaults it sets.

For developers who run Claude Code or Codex in more than one repository and want the same rules, skills and dev container setup in each.

Install

Node.js 22 and at least one agent, Claude Code or Codex (full list). Choose the App to review each change in the browser. Choose the CLI in a terminal-only session or in a script.

App

npx [email protected]

Opens the browser on the Dashboard with the doctor findings (the read-only checks of this machine). Select Fix everything, review the changes and press Apply; nothing is written before.

CLI

npx [email protected] plan

Lists each change and writes nothing.

npx [email protected] apply

Asks per change. --yes applies the default selection without questions.

After either path, restart the agents. npx [email protected] history lists what changed, npx [email protected] undo <run> reverses a run.

Lite level, the plugin alone. Full setup description. What the app writes and what it never does.

If npx cannot find that version, run the app from a clone: steps. Linux and macOS are tested. Windows runs only a CI smoke test.

Every change is shown first

Fix everything shows each file change as a diff and each command as its command line, and writes nothing until you press Apply. The screenshot uses example data. Every view of the app.

More

• Project site: the workflow, the install levels and the app
• Setup guide: prerequisites, levels, what a full apply writes, per-repository setup, updates, undo
• Upgrading to 7.0: clean uninstall of 6.x and clean install of 7, what undo leaves, renamed commands
• Accounts: several Claude Code and Codex logins, signed in through private browser windows, on the host and in dev containers
• Stack: components, repository map, how the stack was chosen, optional tools
• App reference: every command and the web UI
• Automations: scheduled agent runs from templates, installed disabled on the runner from the app or the CLI, with the catalogue, the file format and the lint rules
• Evaluations: the evidence for each component BenchKeeper ships or left out, kept in BenchKeeper-research
• Security notes and how to report a vulnerability
• Changelog and issues

License

MIT