| name | start-local |
| description | Onboarding for OpenAI-compatible local LLM workflows: read STUDIO.md + local-llm skill, then mirror /start routing without Claude Code Slash UI. |
| argument-hint | [no args] |
| user-invocable | true |
| allowed-tools | Read, Glob, Grep, AskUserQuestion |
When to use
User is driving the project from Cursor (or another IDE) with a local Chat Completions–compatible endpoint instead of Claude Code’s slash menu.
Workflow
- Ask the user to confirm
baseURL and model (or read from their docs/ / .env.example if you add one later).
- Read
STUDIO.md and .claude/skills/local-llm/SKILL.md; summarize constraints (no fictional wx/tt, file-write approval, task chunking).
- Optionally run the same discovery steps as
.claude/skills/start/SKILL.md (game concept presence, src/ layout) without assuming /slash exists—refer to skill names as next actions:
- Ideation → skill
brainstorm
- Web stack → skill
setup-web
- WeChat →
setup-wechat-minigame
- Douyin →
setup-douyin-minigame
- Ongoing delivery →
sprint-plan, code-review, etc.
- Never auto-run another skill; let the user invoke the next step explicitly.
Command IDs
| Intent | Skill folder |
|---|
| Local endpoint setup | local-llm |
| First onboarding (full) | start |
| This local-first wrapper | start-local |
