## Description Following the same pattern as the `setup-guardrails` skill (#3586561), the AI module would benefit from a companion skill that configures the module's **general settings** on a running site, driven through the [Tool module](https://www.drupal.org/project/tool)'s tools via `drush tool:run`. Today these settings can only be inspected and changed through the admin UI at `/admin/config/ai/settings`. This task adds a `configure-ai-settings` skill (sibling of `create-guardrail-plugin` and `setup-guardrails`, under `ai/.agents/skills/`) covering the core operational settings stored in `ai.settings`. As with the guardrails skill, the generic `tool_belt` entity tools are content-entity only and cannot read or write `ai.settings` simple config, so dedicated tool plugins are needed so the skill performs each action as a well-defined, access-checked tool invocation. Because the Tool module is **not** a dependency of the AI module, the skill must enable it with drush when it is not already enabled, and uninstall it again afterward only if the skill enabled it, leaving the site's module state unchanged. ## Tasks * [ ] Add the `configure-ai-settings` skill under `ai/.agents/skills/configure-ai-settings/`. * [ ] Provide the settings tool plugins the skill invokes via `drush tool:run`. * [ ] **Default operation-type models** — get and set the per-operation default provider/model (`ai.settings:default_providers`), including the default vector database provider. * [ ] **Providers and models** — get the available AI providers and their models in general (discovered via the `ai.provider` plugin manager). * [ ] **HTTP request timeout** — get and set `ai.settings:request_timeout` (integer seconds, valid range 1–3600). * [ ] **Trusted domains** — get and remove trusted domains (`ai.settings:allowed_hosts`). Adding a trusted domain is intentionally **out of scope for automated writes** (security-sensitive); the skill instead gives the user instructions for adding one through the UI. * [ ] Enable the Tool module via drush when it is disabled, and uninstall it again only if the skill enabled it. ## Acceptance criteria * The default model for any operation type — including the vector database provider — can be read and changed without using the admin UI. * The available providers and their models can be listed. * The HTTP request timeout can be read and set within the allowed range. * Trusted domains can be listed and removed; adding one is not performed automatically — the skill returns clear UI instructions instead. * The site's enabled-module state is identical before and after a skill run (the Tool module is restored to its prior state). ## Related issues Relates to #3586561 (the `setup-guardrails` skill — same architecture and tool-driven pattern).