Agent plugins

wrapido ships one skill, wrapido (skills/wrapido/SKILL.md), packaged as a plugin for several AI coding tools. The skill teaches the agent to:

• start in the wp/v2 namespace and discover the API before acting;
• always run in agent mode;
• authenticate with Application Passwords or OAuth2, and walk you through setting one up when a request needs it;
• fetch every item with --per_page=-1;
• make bulk changes with several ids (batched through /batch/v1) and read partial failures;
• upload files;
• keep output small;
• follow the safety rules (drafts by default, confirm before deleting) and avoid the common gotchas.

The skill is only knowledge: the agent still runs the real wrapido command, so install the CLI first. See AI agents for the full setup, including credentials and a site.

Supported tools

Tool	Agent mode	Guide
Claude Code	Detected (CLAUDECODE)	Claude Code
GitHub Copilot CLI	Detected (COPILOT_AGENT, COPILOT_ALLOW_ALL)	GitHub Copilot
VS Code (Copilot chat)	Not detected, so set WRAPIDO_AGENT=1	GitHub Copilot
OpenAI Codex	Detected (CODEX_*)	Codex
Gemini CLI	Not detected, so set WRAPIDO_AGENT=1	Gemini CLI
Cursor	Detected (CURSOR_AGENT)	Cursor

Where agent mode isn't detected, the skill also tells the agent to prefix its commands with WRAPIDO_AGENT=1 when wrapido --debug --help reports agent mode: off. In any tool, wrapido --debug --help shows whether it's on.

Other tools

Most other agents (OpenCode, Cline, Windsurf, Roo Code, Junie, Amp, Goose and more) read Agent Skills from a skills folder without a plugin. Copy skills/wrapido into that tool's skills folder (often .agents/skills/), or point the agent at AGENTS.md, which is the same guidance as a plain file. Set WRAPIDO_AGENT=1 unless the tool is listed under Agent mode → Turn it on.

How the plugins work

Every tool loads the same skill from skills/wrapido/SKILL.md, which follows the open Agent Skills format. Each tool reads a small manifest that names the plugin and points it at that folder:

Tool	Manifest it reads
Claude Code	.claude-plugin/plugin.json + marketplace.json. This repository is a plugin and its own single-plugin marketplace.
GitHub Copilot (CLI and VS Code)	The Claude Code files, read directly.
OpenAI Codex	.claude-plugin/marketplace.json, plus a generated .codex-plugin/plugin.json that adds a display name, a category and example prompts for its plugin browser.
Gemini CLI	gemini-extension.json (generated).
Cursor	.cursor-plugin/plugin.json (generated).

A skill is loaded only when a request is relevant (anything about a WordPress site), so it costs nothing in other conversations. Because every tool shares one skill, an update reaches all of them at once: update the plugin in your tool (each guide shows how) to pick it up.

For maintainers

.claude-plugin/plugin.json is the single source of truth.

• Edit only that file and skills/. Then run npm run sync:plugins to regenerate gemini-extension.json, .cursor-plugin/plugin.json and .codex-plugin/plugin.json from it. The Codex/Cursor-only extras (logo, example prompts) live in scripts/sync-agent-plugins.js.
• Keep version in .claude-plugin/plugin.json equal to package.json's. You don't bump it by hand for a release: publishing a GitHub Release with tag vX.Y.Z makes release.yml set package.json, package-lock.json and every manifest to X.Y.Z and commit that. Claude Code only offers an update when the version changes.
• CI runs npm run check:plugins, which fails if any of these is true:
  • a generated file is stale;
  • the two versions differ;
  • a skill breaks the Agent Skills name/description rules (VS Code silently drops a skill whose name is invalid).
• CI also runs claude plugin validate ..
• Never add a plugin.json at the repository root. Copilot (and Codex, given a $schema) would prefer it over .claude-plugin/.