[wp-trac] [WordPress Trac] #64769: Backport WP AI Client enhancement to harden security for using `Ability_Function_Resolver`

[WordPress Trac]

#64769: Backport WP AI Client enhancement to harden security for using
`Ability_Function_Resolver`
----------------------------+------------------------------------------
 Reporter:  flixos90        |      Owner:  flixos90
     Type:  task (blessed)  |     Status:  assigned
 Priority:  normal          |  Milestone:  7.0
Component:  AI              |    Version:
 Severity:  normal          |   Keywords:  needs-patch needs-unit-tests
  Focuses:                  |
----------------------------+------------------------------------------
 See https://github.com/WordPress/wp-ai-client/pull/61:

 > This addresses a security weakness: At no point so far we were checking
 (or at least encouraging) that the abilities called in a message are
 actually among the abilities allowed for the prompt.
 >
 > Not checking this can lead to security vulnerabilities, e.g. through
 prompt injection.
 >
 > While developers today could work around this by manually checking prior
 to using `Ability_Function_Resolver`, this is not intuitive at all. Since
 it's a security concern, it needs to be baked in and mandatory.
 >
 > **This is a breaking change.** The `Ability_Function_Resolver` class
 becomes (mostly) non-static, and will require an instance for usage going
 forward. It will require passing the list of abilities in the constructor,
 in the same shape it's used on `Prompt_Builder::using_abilities()`.

 **This is ''not'' an active security vulnerability.** It has been handled
 in public so far, and is reasonable to do so, since it merely improves an
 API to prevent potential future security issues that could have occurred
 with the prior API design.

-- 
Ticket URL: <https://core.trac.wordpress.org/ticket/64769>
WordPress Trac <https://core.trac.wordpress.org/>
WordPress publishing platform