Coding Agents

Docker Doctor is built to work with AI coding agents. There are two pieces: an installable agent skill that teaches your agent the /docker-doctor triage workflow, and a post-scan handoff that sends the issues Docker Doctor just found straight to an agent on your machine.

Install the skill

npx @docker-doctor/cli@latest install

In a terminal this opens a multi-select of coding agents — agents already detected on your machine are pre-selected — and copies the skill into each agent's project-level skills directory (for example .claude/skills/docker-doctor/ for Claude Code).

Works with Claude Code, Cursor, Codex, OpenCode, and every other agent supported by agent-install.

Non-interactive install

In scripts or CI, pass the agent ids explicitly:

npx @docker-doctor/cli@latest install --agent claude-code cursor

Any agent-install agent id is accepted. Unknown ids fail with the full list of valid ones.

Global install

Pass --global to install the skill once for your whole machine instead of the current project — into each agent's global skills directory (for example ~/.claude/skills/ for Claude Code):

npx @docker-doctor/cli@latest install --global --agent claude-code cursor

This is also what the Docker Sandboxes kit uses inside sandboxes, so the mounted workspace stays clean.

Install as a plugin

If you prefer your agent's plugin system over copied skill files, the PunGrumpy/docker-doctor repository doubles as a plugin for both Claude Code and Cursor. A plugin install bundles all three skills — docker-doctor (scan and triage), docker-author (write production-grade Dockerfiles), and improve-docker (audit and roadmap) — and updates when the repo does, instead of freezing a copy at install time.

Claude Code

The repository is a Claude Code plugin marketplace. Inside Claude Code:

/plugin marketplace add PunGrumpy/docker-doctor
/plugin install docker-doctor@docker-doctor

Cursor

The repository ships a Cursor plugin manifest (.cursor-plugin/plugin.json), so it can be added as a plugin from the repo URL wherever Cursor accepts one.

Plugins and install copies coexist fine; the skill content is identical. Pick one to avoid your agent seeing duplicate skills.

Hand issues to an agent after a scan

When an interactive scan finds issues, Docker Doctor asks what to do next:

✔ What would you like to do next?
❯ Claude Code
  Copy prompt to clipboard
  Skip

• <agent name> — shown for each agent CLI found on your PATH (claude, codex, cursor-agent). Docker Doctor installs the skill for that agent, then launches it in your terminal with a prompt describing every issue, its fix, and the affected files. The agent runs in its auto-approve mode so it can fix the issues end-to-end.
• Copy prompt to clipboard — copies the same prompt so you can paste it into any agent or chat.
• Skip — do nothing.

The .docker-doctor/ report directory

Before handing off, Docker Doctor writes the full scan results to .docker-doctor/ in your project root:

• diagnostics.json — the same report as --json
• one .txt per rule with the fix recipe and every affected file

The prompt references this directory so the agent can read past the inline summary. Docker Doctor adds .docker-doctor/ to your .gitignore automatically (only if the project is a git repository and the entry is missing).