Summary

The Console is shipping a guided wizard (use-case → language → best-match template → where the source lives). We want the same guided experience in the CLI, so that developers and AI agents are led through Actor creation, and so a user who starts in the Console can hit "Clone locally" and continue seamlessly in the terminal.

This reshapes the existing apify create command rather than adding a new one.

Console wizard being mirrored

1. What do you want to build? — Web scraper · API & data pipeline · AI agent · Skip (show all)
2. Which language do you prefer? — JavaScript · TypeScript · Python · Other (bring-your-own runtime via Docker)
3. Here's your best match — recommended template(s)
4. Where will the source code live? — Apify · Connect GitHub/GitLab/Bitbucket (Apify creates the repo, builds on every push)
5. Confirm & install

Guiding principle

The wizard is an interactive flag-filler. Every step maps to exactly one flag. Humans answer prompts, agents pass flags, and the Console's "Clone locally" button emits the identical apify create … line. No flag is wizard-only; no step lacks a flag.

This single rule gives us human, agent, and Console-handoff support from one surface.

Reshaped apify create wizard (order mirrors the Console)

apify create
 → Name your Actor                  [actorName arg]
 → What are you building?           [-u, --use-case]   web-scraping · api-pipeline · ai-agent · skip
 → Which language?                  [-l, --language]   JavaScript · TypeScript · Python
 → Here's your best match           [-t, --template]   recommended first; SKIP this prompt if the filter leaves exactly one
 → Where will the source live?      [--source]         Apify · GitHub · GitLab · Bitbucket
   └─ (provider) → git connect flow
 → confirm / install

@inquirer/select per-choice description lines give us the Console's card subtitles (e.g. "Crawlee + Node.js. The most popular runtime on Apify.").

Flag surface

Three consumers, one command

• Human: apify create, answer prompts.
• Agent: apify create acme -u web-scraping -l python -t python-crawlee --source apify --yes --json. Discovers valid enums via apify templates ls --json (exposes useCases[] + recommended) — never scrapes --help.
• Console "Clone locally": emits that exact line with --origin console.

Git flow (--source <provider>) — resolved

The local scaffold is the source of truth; the Apify platform creates the repo from it, browser only for one-time provider auth.

1. collect choices
2. scaffold template → git init → initial commit            ← local files exist immediately
3. ensure provider authorized → open browser to verify/authorize (one-time only)
4. Apify platform creates the repo on the connected provider,
   creates the Actor (sourceType=GIT_REPO, gitRepoUrl), returns the remote URL
5. git remote add origin <url> → git push -u origin main
6. ✔ connected — Apify rebuilds on every push

• Provider-agnostic: the browser does the auth, so --source github|gitlab|bitbucket just picks which connect URL to open — no per-provider code in the CLI.
• Agent-safe: provider already authorized + no TTY → steps 3–5 run fully non-interactively. Not authorized → emit gitConnectUrl / nextSteps[] in --json and never block on a browser.

Notes & scope boundaries

• "Import from Git" (Console's top-right button) = register an existing repo, not template scaffolding → tracked separately as apify actor connect-git / apify import. Out of scope here.