Abilities API WordPress.org Plugin

[Jameswlepage]

Publish a temporary, experimental “Abilities API” plugin on WordPress.org to maximize testing/feedback ahead of core inclusion; close it after merge.

What it is
-Bundled Abilities API (no Composer needed on target sites), though perhaps it could use Composer to serve as an example of how you would bundle abilities into a plug in before the merge
-Small demo set of abilities (safe/read‑first; no destructive actions), the same that will eventually make it to Core, for early testing and exploration

Clear messaging around the plugin being stable but experimental and tooling to make it easy for GitHub releases to update the. org plugin as well is needed.

Is there anything else that we should be thinking about before going ahead with this?

[Jameswlepage]

Related: WordPress/ai#40 - If we create these abilities to include in this plugin, we can get some real world feedback on if this list makes sense.

[gziolo]

I was wondering whether AI Experiments Plugin shouldn't play the role of showcasing how the Abilities API could be used in combination with other work, like PHP AI SDK, similar to what @felixarntz built in https://github.com/felixarntz/wp-ai-sdk-chatbot-demo for the WC US demo. We could still develop a few baseline abilities here, but my primary concern is how to optimize the experience for those who want to test all the AI efforts efficiently, because my biggest worry is that these default abilities alone do nothing until there is a glue code that integrates them with the rest of WordPress.

[justlevine]

I too assumed that the AI Experiments would be the (eventual) holder for Abilities before they graduate their way into core, and that the feedback we were looking to gather for the quick merge path was API ergonomics, not the functionality/usefulness of individual Abilities.

(Middle ground is to keep them in this plugin until we're ready to merge in order to help with "discovery", and then triage them either into core or into AI experiments)

Is there anything else that we should be thinking about before going ahead with this?

The type/sort of feedback we're looking to gather.

Answering this would give us a better litmus for:

• which abilities to include here versus downstream.
• whether the Composer false premise is relevant here, and more broadly what expected/recommended steps implementers need to "bundle abilities into a plug in before the merge"

---

From a technical POV we already have things set up to build a working plugin distribution (no end-user Composer necessary: https://github.com/WordPress/mcp-adapter/blob/trunk/.wp-env.json#L4 ), so only thing required from CI/CI is the push to .org

[felixarntz]

I think developing abilities for exploration in the Abilities API plugin would make sense. For the 6.9 Core merge, I would suggest that we only merge the API itself though as I believe it takes more time to come up with a cohesive set of abilities that demonstrate real value against different product/project needs.

This model would follow the REST API, which was added to WordPress Core without any actual endpoints built-in, those came a bit later.

I do think we should start developing initial abilities in the plugin ASAP, even if we don't want to merge them into 6.9, simply to dogfood the API and ensure it satisfies all crucial needs for implementing abilities. For example, in my work on https://github.com/felixarntz/wp-ai-sdk-chatbot-demo I already identified a few issues (see #53), so that shows how crucial building against it early on is.

It would also establish the Abilities API plugin as a long-term development project: First priority is to build out the API to get it into Core (what we're doing now), second priority is to develop abilities and keep it around as a long-term canonical plugin, with sets of capabilities that are considered mature and useful enough being merged from the plugin into Core iteratively.

[gziolo]

It would also establish the Abilities API plugin as a long-term development project: First priority is to build out the API to get it into Core (what we're doing now), second priority is to develop abilities and keep it around as a long-term canonical plugin, with sets of capabilities that are considered mature and useful enough being merged from the plugin into Core iteratively.

In that scenario, we could still leverage the AI Experiments plugin to integrate these general-purpose abilities, deeply showing its full potential. I can see how that arrangement might work in the long run.

For development purposes, it would probably be easier to do all the experiments first inside the AI Experiments plugin, but I'm sure we can find an interesting way to test it regardless while shipping in the Abilities API canonical plugin.