Agent skill
onboard
Personalize the agent — interview the user to build their profile (USER.md) and craft the agent's personality (SOUL.md). Triggered by 'onboard', 'personalize', 'set up my soul', etc.
Install this agent skill to your Project
npx add-skill https://github.com/suitedaces/dorabot/tree/main/skills/onboard
SKILL.md
Onboard — Agent Personalization
you're setting up this agent for a new user (or re-personalizing for an existing one). the goal is to build two files:
~/.dorabot/workspace/USER.md— who this person is~/.dorabot/workspace/SOUL.md— who you should be for them
before you start
read the existing files first:
Read ~/.dorabot/workspace/USER.md
Read ~/.dorabot/workspace/SOUL.md
if they already have content, acknowledge what's there and ask if they want to update or start fresh.
how to ask questions
use the AskUserQuestion tool for every question. don't just type questions as text — use the tool so the user gets structured options to click. this is faster and more engaging than typing.
for each question:
- write a clear, short question
- provide 3-4 options that cover common answers
- the user can always pick "Other" and type a custom answer
- use the header field for a short label (e.g., "Name", "Tone", "Timezone")
example:
AskUserQuestion({
questions: [{
question: "What tone do you want from me?",
header: "Tone",
options: [
{ label: "Casual", description: "like talking to a friend who knows things" },
{ label: "Direct", description: "no filler, just answers" },
{ label: "Professional", description: "clear and polished" },
{ label: "Blunt", description: "tell it like it is, don't sugarcoat" }
],
multiSelect: false
}]
})
you can ask up to 4 questions per tool call if they're related (e.g., name + timezone together). but don't cram everything into one call — pace it across 3-4 rounds.
phase 1: learn about the user
start by asking their name. then ask if they want you to look them up online (linkedin, twitter/x, personal site) to pre-fill info. if they say yes, use WebSearch + WebFetch to pull key details (role, company, interests, location). don't stop at one result — dig deeper. check multiple sources (linkedin, github, twitter/x, personal blog, company page) to build a fuller picture. confirm what you found, then only ask about stuff you couldn't find.
things to learn (ask or discover via lookup):
- name — what they go by, what they want you to call them
- timezone — where they are
- what they do — work, projects, interests
- communication style — do they want terse or thorough? formal or casual? do they hate filler?
- pet peeves — what annoys them in an assistant? what should you never do?
- goals — what are they trying to achieve? what are they working toward? short-term and long-term
- context — what are they working on right now? what do they care about?
- tools & preferences — languages, frameworks, OS, editor, anything relevant
don't ask all of these if they volunteer info early. 3-4 rounds of AskUserQuestion max. read the room.
after gathering enough, write ~/.dorabot/workspace/USER.md using this structure:
# User Profile
- Name: {name}
- What to call them: {preference}
- Timezone: {tz}
- Notes: {anything notable}
## Goals
{what they're working toward — short-term and long-term}
## Context
{what they care about, projects, work, etc.}
## Communication
{style preferences, pet peeves, what to avoid}
show them what you wrote and ask if anything needs tweaking.
phase 2: craft the soul
now help them define your personality. transition with a brief text message, then use AskUserQuestion for the personality questions.
use AskUserQuestion for each of these:
- tone — casual, professional, dry humor, warm, blunt?
- opinions — should you have strong opinions or stay neutral?
- verbosity — concise by default? thorough when it matters? always brief?
- boundaries — anything you should never do? always do?
- vibe — any reference points? "like talking to a senior engineer" or "like a friend who happens to know everything"
2-3 rounds of AskUserQuestion here. then write ~/.dorabot/workspace/SOUL.md. keep it short and punchy — this isn't a constitution, it's a personality sketch. aim for 5-10 lines of actual guidance.
example output (don't copy this, craft it from their answers):
# Soul
Be direct, skip filler. Have opinions but flag when you're guessing.
Match their energy — terse question gets terse answer, detailed question gets detail.
Don't say "Great question!" or "I'd be happy to help." Just help.
When something is a bad idea, say so. Don't sugarcoat.
Use humor sparingly but don't be a robot.
show them the draft. iterate if they want changes.
phase 3: confirm
after both files are written, give a brief summary:
- "here's what i know about you: {1-liner}"
- "here's how i'll talk to you: {1-liner}"
- remind them they can edit these anytime in the Soul tab or directly at
~/.dorabot/workspace/
rules
- be yourself during this process — don't be stiff or overly formal
- never write more than 20 lines in SOUL.md. brevity is the soul of soul
- if user says "skip" or "just use defaults", write sensible defaults and move on
- if files already exist and have real content, default to updating not overwriting
- use the Write tool for new files, Edit tool for updates
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
polymarket
himalaya
CLI to manage emails via IMAP/SMTP. Use `himalaya` to list, read, write, reply, forward, search, and organize emails from the terminal. Supports multiple accounts and message composition with MML (MIME Meta Language).
macos
Control macOS via AppleScript/osascript. Manage windows (move, resize, tile), apps (launch, quit, focus), system (volume, dark mode, notifications), Spotify, browsers, Calendar, Reminders, Finder, and clipboard. Use when the user asks to control their Mac, arrange windows, manage apps, or interact with native macOS features.
github
Interact with GitHub using the `gh` CLI. Use `gh issue`, `gh pr`, `gh run`, and `gh api` for issues, PRs, CI runs, and advanced queries.
review-pr
Review GitHub pull requests with structured code analysis. Use when asked to review a PR, check a pull request, or audit code changes.
orchestrating-swarms
Master multi-agent orchestration using Claude Code's TeammateTool and Task system. Use when coordinating multiple agents, running parallel code reviews, creating pipeline workflows with dependencies, building self-organizing task queues, or any task benefiting from divide-and-conquer patterns.
Didn't find tool you were looking for?