Add studio ai CLI command with interactive AI agent

[youknowriad]

Related issues

• Related to AI assistant exploration for Studio CLI

Proposed Changes

• Add a new studio ai command that launches an interactive AI-powered WordPress assistant
• Integrate the Claude Agent SDK with a terminal UI (pi-tui) for rich markdown chat
• Provide MCP tools for site management: listing sites, getting site info, starting/stopping sites, running WP-CLI commands
• Bundle WordPress development skills as SDK plugins (creating/editing blocks, themes, and patterns)
• BYOK (Bring Your Own Key) authentication with API key persistence in appdata
• Session resume across conversation turns within the same REPL session
• Escape key to interrupt the agent mid-turn without losing context
• Human-readable tool status in the spinner (e.g., "Running WP-CLI" instead of mcp__studio__wp_cli)
• Tool result output displayed inline with dimmed formatting

Testing Instructions

Prerequisites

• An Anthropic API key (get one at https://console.anthropic.com)
• At least one WordPress site created in Studio

Setup

1. npm install
2. npm run cli:build

Basic flow

1. Run node dist/cli/main.js ai
2. On first run, you'll be prompted for your Anthropic API key — it will be saved for future use
3. The chat interface should appear with a bordered editor at the bottom

Test conversations

Try these prompts to exercise different capabilities:

Site management tools:

• "List all my WordPress sites" — should call site_list and display results
• "Start [site-name]" — should start the site and confirm
• "What plugins are installed on [site-name]?" — should use wp_cli with plugin list

WordPress development skills:

• "Create a custom WordPress block plugin called 'testimonial-card' that displays a quote, author name, and photo" — should load the creating-blocks skill and write files
• "Create a minimal block theme called 'minimal-starter'" — should load the creating-themes skill

UI behavior:

• Verify the spinner shows human-readable labels (e.g., "Listing sites", "Reading file", "Thinking…")
• Verify tool results appear as dimmed ↳ lines below the spinner
• Verify error results appear with red ✗ prefix
• Verify text from the agent has line breaks between consecutive messages (not concatenated)
• Press Escape during a long operation — should interrupt the agent gracefully
• Press Ctrl+C — should exit cleanly
• When the agent asks questions (AskUserQuestion), verify options are numbered and you can type a number to select

Session continuity:

• After one prompt completes, send a follow-up that references the first (e.g., "now add a settings panel to that block") — should resume the session context

Edge cases:

• Try asking about a site that doesn't exist — should get a clear error
• Try running WP-CLI on a stopped site — should tell you to start it first
• Run with --max-turns 5 and give a complex task — should hit the limit with a clear message

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?

[youknowriad]

This is not just a Proof of concept in my mind but yeah still early :P

[wojtekn]

This is not just a Proof of concept in my mind but yeah still early :P

I used 'Proof of Concept' to signal that we haven't settled on the exact approach yet, not to say the ideas aren't valuable - they are! We're actually about to start a project to integrate Studio with ACP, and what you've built here will definitely help shape that work. Since we don't know the final architecture yet, I'd rather not merge something we might need to rewrite later - but I'd love to keep this PR as a reference as we figure out the direction. Would that work for you?

[youknowriad]

Let's discuss a bit after but IMO, We should ship something if it's valuable. I agree with not merging the PR before it's ready though if that's what you meant but if you see this PR as just a "vision PR", this is not my intent, my ultimate intent is to ship the PR.