onboard

name	onboard
description	Lightweight first-run interview that seeds the ongoing-context-capture pages and hands off to /help
risk_class	soft

Onboard

A short first-run interview. The goal is the smallest viable amount of context the Agent needs to be useful in the next conversation — not a comprehensive intake. Heavy context-building is delegated to /ingest-writing, which the user can invoke any time with raw material.

When to invoke

• The first time the user opens a vault — check rubber-ducky settings get onboard.completed. If it returns false (the default), proceed with the interview. If it returns true, skip silently unless the user explicitly asks.
• Whenever the user explicitly asks: "onboard me", "help me get set up", "introduce yourself", "what should I tell you about me?"

If onboard.completed is already true, don't re-run silently — confirm: "We've onboarded before — re-run, top up just one section, or skip?" This flag is the authoritative "have we onboarded?" state. The wiki log is append-only history; the flag is current state.

Interview shape

One question per turn, prefixed with its position so the user knows the commitment up front. The whole interview is 4 questions:

Question 1 of 4. In one sentence, what's this vault going to hold? (Work, a project, your boat, your personal life, anything.)

Question 2 of 4. What should I call you, and is there anything about your role I should know to make my replies useful?

Question 3 of 4. Are there any internal-system names, acronyms, or in-house words that come up a lot? (e.g., project codenames, tool names.)

Question 4 of 4. Any communication preferences? Short or long answers, casual or formal, anything I should avoid?

Skip Q3 if the user signals the vault is personal / non-work, and skip Q4 if the user already gave a clear style signal in Q2.

Persistence

After each answered question, route the answer through /ingest-writing so it lands in the right page (don't write to the context pages directly):

Question	Page
1	wiki/about.md — Projects section, prefixed with the vault's purpose
2	wiki/about.md — Identity section
3	wiki/vocabulary.md
4	wiki/preferences.md

Audit and mark completion once the interview is done:

rubber-ducky log append "[onboard] completed lightweight intake (4 questions)"
rubber-ducky settings enable onboard.completed

The settings flip is the authoritative signal that future invocations read on entry; the log entry is the human-readable trail.

Paste-raw invitation

Before the hand-off, check the auto_on_onboard setting:

rubber-ducky settings get ingest.auto_on_onboard

If it returns true (the default), invite the user to paste raw material:

Last thing — paste any writing you already have about yourself or what this vault is for. Old bios, performance reviews, articles, project docs, anything. I'll route what's useful into the right pages. (Or just say "skip" — you can paste any time later with /ingest-writing.)

If the user pastes, hand the raw text straight to /ingest-writing (skill-to-skill). If they skip, move on.

If auto_on_onboard is false, skip the invitation entirely.

Hand-off

End the session with the canonical "now what?" pointer:

I'm rubber-ducky. Type help anytime to see everything I can do.

Do not dump the full capability list here — that's /help's job. Keep this final line a single sentence so the user lands in a clean state and decides their own next move.

What this skill does NOT do

• No exhaustive interview. Four questions, max. The plan explicitly treats context capture as ongoing — anything missed here gets captured later via paste-raw.
• No backend wiring. Integrations are added on demand via /connect. Don't ask about Jira / Slack / GitHub here.
• No project / task creation. This skill seeds context only. The user describes work to the Agent in normal conversation afterward.