setup-walnut

# Set up Open Walnut — we do it for you **Goal:** make Open Walnut's main agent (the "Personal AI") able to talk on first launch. You are doing the setup *for* the user. Be **proactive**: do everything you can yourself, write the config yourself, verify with a real model call yourself, and fix errors yourself. Ask the user a question **only** when you genuinely cannot proceed without an answer they alone have. ## Two truths that shape this skill 1. **The Personal AI needs a real model credential.** Walnut's main agent calls the model API **directly via the SDK** (Bedrock / Anthropic / OpenAI / Google / Ollama). It does **not** go through the `claude` CLI and has no OAuth path. So the Personal AI (home-page chat, triage, memory, cron) only works once one real credential is resolvable. 2. **Coding sessions already work with zero config.** The `/sessions` Claude Code sessions spawn the user's **own logged-in `claude` binary** and use its login — they need *nothing* from Walnut's config. So even before the Personal AI has a credential, the user can clone, start, and open coding sessions. Tell them this so they're never blocked: *"You can start using coding sessions right now; let's also wire up the Personal AI so home-page chat works."* Walnut resolves the Personal AI's Bedrock credential with this priority: `~/.open-walnut/config.yaml` → `~/.claude/settings.json` env block → process env → `~/.aws`. Writing `config.yaml` is the most explicit and portable...