Skip to content

Claude Code

The wrapido skill installs as a plugin in Claude Code. It teaches Claude to drive the wrapido CLI: discover the site's REST API first, run in agent mode, authenticate, page through everything, make bulk changes, upload files, and follow the safety rules.

How it works

This repository is a Claude Code plugin and its own single-plugin marketplace (.claude-plugin/plugin.json and .claude-plugin/marketplace.json). The plugin's only content is the skill in skills/wrapido/SKILL.md, which Claude loads when a request is about a WordPress site.

1. Install the CLI

Claude runs wrapido from your shell, so it must be on your PATH. See Installation, then check with wrapido --help.

2. Install the plugin

claude plugin marketplace add spacedmonkey/wrapido
claude plugin install wrapido@wrapido

You can also run /plugin inside Claude Code. To pick up later changes:

claude plugin marketplace update wrapido

To try a local checkout instead of GitHub, add the folder: claude plugin marketplace add ~/path/to/wrapido.

Without the plugin

Claude Code also loads plain skill folders. For every project on your machine:

mkdir -p ~/.claude/skills
cp -r skills/wrapido ~/.claude/skills/

For a single project, copy it to that project's .claude/skills/wrapido/ instead.

Restart Claude Code (or start a new session) afterwards. Run /skills to confirm wrapido is listed.

3. Agent mode

Agent mode switches on by itself inside Claude Code, so there is nothing to configure. Check with wrapido --debug --help, which prints agent mode: on (CLAUDECODE). See Agent mode.

4. Credentials and a site

Export WP_USERNAME/WP_PASSWORD (an Application Password) before starting Claude Code, and put url: https://example.com in a wrapido.yml in your project. See AI agents → Credentials and a site for OAuth2 and other options, including the warning about wrapido.yml in cloned repositories. If you skip this, Claude walks you through it the first time a request needs auth.

5. Use it

Claude loads the skill on its own when you ask for something WordPress-related. You can also invoke it directly with /wrapido:

  • "List the five most recent draft posts."
  • "Create a draft post titled Hello World with this content: ..."
  • "Upload ./cat.jpg to the media library and set its alt text."
  • "Which custom post types and taxonomies does this site register?"
  • "Find posts missing a featured image."
  • "Get the titles of all published posts." (uses list --per_page=-1)
  • "Upload ./logo.png and set it as post 12's featured image."

Claude goes straight to the wp/v2 namespace for core content and will typically run wrapido help <namespace> <route> <verb> --format=json first to read the live schema, then run the real command with --fields to keep the output small. It creates content as draft, shows what will change before an update or delete, and asks before bulk or irreversible work.

Permissions

Plugins can't grant permissions, so Claude Code prompts before every wrapido call by default. To allow the read-only commands without a prompt, add this to .claude/settings.json (one project) or ~/.claude/settings.json (everywhere). Writes (create, update, delete, generate, meta) stay on a prompt:

{
    "permissions": {
        "allow": [
            "Bash(wrapido help:*)",
            "Bash(wrapido config get:*)",
            "Bash(wrapido * list:*)",
            "Bash(wrapido * get:*)",
            "Bash(wrapido * exists:*)"
        ]
    }
}

Patterns match the command text, so a read with flags before the verb (for example wrapido --url=... wp/v2 posts list) will still prompt.