Skill: First-Run Welcome

Purpose

Provide an adaptive welcome experience based on setup state. Routes new users through

/setup

for guided onboarding. Welcomes returning users with context about their active dataset and quick actions.

When to Use

• Session start (triggered by Knowledge Bootstrap)
• Before any analysis work begins

Instructions

Step 1: Detect setup state

Read

.knowledge/setup-state.yaml

. Classify into one of three states:

1. Cold start — file does not exist OR

   setup_complete: false

   with no

   phases_completed

   (empty or missing).
2. Partial setup — file exists,

   setup_complete: false

   , and at least one entry in

   phases_completed

   .
3. Warm start — file exists and

   setup_complete: true

   .

Step 2: Route based on state

---

Cold Start (no setup-state.yaml or setup_complete: false, no phases done)

Present this welcome and route to

/setup

:

Welcome to AI Analyst — your analytical partner for product teams.

I help you turn business questions into validated insights, charts, and
presentations. Think funnel analysis, segmentation, root cause investigation,
trend detection — from question to slide deck.

Let's get you set up. I'll walk you through a quick interview to learn about
your data, your role, and what you want to analyze.

Starting setup now...

Then invoke

/setup

to begin the guided interview. Do NOT show dataset info, tutorial content, or example queries. The setup flow handles all onboarding.

---

Partial Setup (some phases complete, setup not finished)

Read

phases_completed

and

phases_remaining

from

.knowledge/setup-state.yaml

.

Welcome back! Your setup is partially complete.

Done: [list phases_completed]
Remaining: [list phases_remaining]

Want to pick up where you left off? Type `/setup` to resume, or ask me
a question if you'd rather dive in.

---

Warm Start (setup_complete: true)

Read context from:

• .knowledge/active.yaml

  →

  active_dataset

  name

• .knowledge/datasets/{active}/manifest.yaml

  → table count

• .knowledge/analyses/index.yaml

  →

  last_updated

  for last analysis date

Welcome back! Here's where things stand:

Dataset: [DATASET_NAME] ([N] tables)
Last analysis: [DATE or "none yet"]

Quick actions:
- Ask a question — "What's our conversion rate by channel?"
- /explore — interactive data exploration
- /run-pipeline — full analysis from question to deck

What would you like to work on?

If

active_dataset

is null (setup complete but no data connected), show:

Welcome back! Setup is complete but no dataset is active yet.

- /connect-data — add a dataset
- /datasets — see available datasets

What would you like to do?

Step 3: Proceed

After presenting the welcome:

• Cold start: Hand off to

  /setup

  . Do not proceed with analysis.
• Partial setup: If user types

  /setup

  , hand off. If user asks a question, route through Question Router and note that setup can be finished later.
• Warm start: If user asks a question, route through Question Router. If user picks a quick action, invoke that skill/agent.

Anti-Patterns

1. Never show welcome to warm-start users who typed a question. If their first message is a question, answer it — weave a one-line "welcome back" naturally.
2. Never show dataset details or tutorial content on cold start. The

   /setup

   flow handles all onboarding.
3. Never overwhelm with feature lists. Keep each welcome variant concise.
4. Never reference NovaMart, bootcamp, or workshop content. This is a general-purpose tool, not a course.
5. Never block on welcome. If the user already asked a question, serve it — adapt the welcome around their intent.