[PROMO] WP-Autoplugin v2 brings real agentic coding inside WP
I've just released WP-Autoplugin v2, a free plugin for building and modifying WordPress plugins with AI.
V1 was essentially an AI code generator. V2 is a complete rewrite and works much more like Claude Code or Codex, except the coding agent runs directly inside wp-admin and understands that it's working with WordPress plugins and themes.
It uses actual agentic tool loops: the model can explore an installed codebase, search/read relevant files, follow hooks and dependencies, make a plan, generate focused changes, review the result, and iterate based on what it finds.
The basic workflow is Plan → Code → Review.
Some things you can do with it:
- Create a complete plugin from a description
- Fix or extend an existing installed plugin/theme
- Have the agent explore an unfamiliar codebase and explain how it works
- Create a separate add-on plugin by inspecting the target plugin's actions/filters
- Review generated changes as diffs and manually edit them
- Iterate on plans and code without losing previous revisions
- Package/download the result, install it as a fork, or apply it to the original plugin/theme
One thing I was particularly careful about: the agent never directly edits the live plugin/theme while it's working. Changes are staged as separate revisions, and modifying the installed code requires explicit approval.
The agent's access is also deliberately limited: it gets bounded, read-only tools for exploring source code rather than unrestricted access to the WordPress installation.
It's completely free (no Pro version, no subscription) and it's BYOK, with support for OpenAI, Anthropic, Gemini, xAI and compatible APIs. You can also use a ChatGPT subscription via its Codex credits, so you don't necessarily need API keys at all.
The whole thing runs through WordPress/PHP (with background jobs for longer agent loops), so there's no Node service or separate coding environment to set up.
GitHub: https://github.com/WP-Autoplugin/wp-autoplugin
I'd love to hear what you think, feedback, bug reports, ideas, and real-world use cases are all very welcome.
[link] [comments]