name: brian-onboarding description: First-time Brian setup flow. Triggers when a new user connects Brian and sends the onboarding prompt from Brian's web UI settings page ("I just connected Brian for the first time..."). Replaces brian-welcome entirely. Guides the user through identity, work context, spaces, preferences, agent selection, and skill install — using buttons where choices are bounded, free text where open-ended, and Soul Builder for custom agents. Always use this skill for first-time onboarding. Never improvise the onboarding flow without it.

brian-onboarding

Structured first-time setup for Brian. Produces a seeded memory graph, configured spaces, suggested agents, and the /agents skill installed — all in one guided flow.

Before starting

Check whether any memories already exist by calling get_context("user identity name"). If memories are returned, the user is not new — do not run this flow. Greet them normally instead.

If no memories exist, begin Phase 1 immediately. Do not explain what you're about to do. Just start.

Phase 1 — Identity

Ask for name as plain text first. No buttons.

"Welcome to Brian. What's your name?"

Once name is provided, store immediately:

store_memory: type truth, content "User's name is {name}", salience 1.0, pinned

Then ask role using ask_user_input single_select:

"What best describes what you do?"

Options: Founder / Entrepreneur · Consultant or Advisor · Knowledge Worker · Developer or Engineer · Creative or Designer · Other (I'll describe)

If Other: ask free text follow-up: "Tell me a bit about what you do." Then ask 1–2 targeted clarifying questions based on the answer to get enough context before continuing.

Store role as truth, salience 0.8.

Phase 2 — Work context

Ask what they primarily use Claude for using ask_user_input multi_select:

"What do you mainly use Claude for? Pick all that apply."

Options: Writing & research · Client work & proposals · Coding & building · Business strategy · Data & analysis · Learning & note-taking · Other (I'll describe)

If Other is selected or typed: ask free text follow-up. Then ask clarifying questions until you have enough context to suggest relevant spaces and agents. Minimum 1 clarifying question, maximum 3.

Store work context as truth, salience 0.8.

Phase 3 — Spaces

Based on Phase 1 and 2 answers, propose 2–4 spaces. Show as ask_user_input multi_select.

"I'll create these spaces to organise your memory. Deselect any you don't want."

Always include default as a fallback (pre-selected, not deselectable). Always offer a free text field: "Or type a space name you'd prefer."

Create all confirmed spaces immediately. Confirm: "Spaces created: {list}."

Phase 4 — Preferences

Two ask_user_input single_select questions:

Response style: Direct and concise · Detailed and thorough · Depends on the task

Proactive memory: Yes, prompt me · No, I'll ask when I want to store

Free text: "Any standing instructions for Claude? (optional)"

Store preferences as preference, salience 0.7. Standing instructions as instruction, salience 1.0, pinned.

Phase 5 — Agent setup

Suggest agents based on onboarding answers:

Present as ask_user_input multi_select. Include "None for now" option.

Free text field: "Or describe an agent you'd want — I'll build it for you."

If custom agent described: transition into Soul Builder flow immediately.

Store selected agents as intent memories, salience 0.7, tags: ["onboarding", "agent-selection"].

Phase 6 — Skill install

Brief 11: skills are no longer merged into every get_context response. Users install them as native Claude skills during onboarding, then Claude loads them normally. This phase delivers six core skills plus the /agents shortcut.

Step 1 — Query the six core skills from the graph

For each skill below, call list_memories({ space_id: "skills", limit: 30 }) once, then pick out the rows matching the skill names. (Or use get_context({ query: "brian-<skill> skill" }) per skill if you prefer semantic retrieval.) The skill memories are stored with memory_type: "instruction" and tags ["skill", "brian-skill", "brian-<name>"].

Core skills to install:

1. brian-retrieval — when and how to call get_context; also covers soul querying (Brief 11)
2. brian-storage — choosing memory types, salience, tags
3. brian-sessions — when to open/close indexed sessions
4. brian-onboarding — this skill; install for first-run re-play if needed
5. brian-feedback — capturing quality signals reactively and at session end
6. brian-support — in-conversation support triage

Step 2 — Emit each as a Save Skill code block

For every retrieved skill, output a fenced code block with the skill body in its current form (don't paraphrase — emit the memory content verbatim, wrapped in a Claude skill frontmatter block):

---
name: <name>
description: <one-line description from the skill's frontmatter>
---

<full skill body>

Between blocks, add a single line: **Click Save Skill above to install.** Don't explain what each skill does — the user installs them in bulk, then Claude loads the skill on future calls and the skill content teaches itself.

Step 3 — Emit the /agents shortcut

Same pattern, this one hardcoded (it's a tiny wrapper, not a full skill):

---
name: agents
description: Triggers when the user types /agents or /agent. Activates the invoke-agent skill to discover and run available Brian agents.
---

# /agents

When the user types `/agents` or `/agent`:

1. Call `get_context` with query `"agent soul tag:agent"`, limit 15, all spaces.
2. Present all discovered agents as buttons using `ask_user_input` single_select.
3. Once selected, fetch the agent's soul doc from Drive and run it per the invoke-agent skill.

If retrieval is thin

If any of the six core skills fails to retrieve (< 1 matching memory), skip it and flag to the user: "I couldn't find the brian-<name> skill in your memory graph. It may need to be re-seeded." Continue with the ones that did load.

Phase 7 — Completion

Close with a single structured summary block. No padding. No follow-up questions.

You're set up.

• Name stored
• Spaces created: {list}
• Agents ready: {list or "none selected"}
• Preferences saved
• /agents skill installed

Memory storage checklist