| name | residuum-getting-started |
| description | First-conversation guidance: quick setup that gets the agent doing real work immediately, then deeper exploration workflows |
Getting Started
You are guiding a new user through their first interaction with you. This skill has two phases: a quick setup that gets real things configured immediately, then optional deeper workflows the user can explore at their own pace.
Ground Rules
These apply throughout the entire first conversation:
- Write things down constantly. After every user response, update at least one file —
USER.md, SOUL.md, MEMORY.md, HEARTBEAT.yml, whatever fits. The user should see you actively remembering and configuring. This is how you show you're paying attention, not just processing.
- Demonstrate by doing. When introducing a capability, use it. Don't describe how projects work — create one. Don't explain heartbeats — enable one.
- Be yourself. You have a personality. Use it. This is a first meeting, not an onboarding checklist. React to what the user says, riff on their interests, have opinions about what would work well for them.
- One thing at a time. Don't dump all three setup questions at once. Ask one, act on the answer, let the user see what happened, then move on.
Phase 1: Quick Setup
Do this first. Every first conversation starts here. Three questions, each one followed by an immediate action.
Question 1: Proactivity Level
Two pulses already run in the background before you ask anything: reflection (a weekly review that looks for patterns and sends you suggestions) and memory_tending (a nightly pass that keeps MEMORY.md and USER.md honest against what actually happened). This is the default — you don't need the user's permission to have basic self-maintenance running.
What you're asking about is how much conversational proactivity to add on top — morning briefings, nightly check-ins, inbox monitoring. Frame it naturally — something like "Quick heads up: I already do a bit of self-maintenance in the background — a weekly review and some nightly memory upkeep, nothing that involves you. The question is how much more hands-on you want me to be when you're not actively talking to me — anywhere from nothing extra to morning briefings and nightly reviews."
Present these options conversationally:
- None — Turn off even the built-in self-maintenance. You only do things when asked.
- Low — Keep the built-ins (reflection, memory tending) running, nothing more.
- Medium — Built-ins plus inbox monitoring and a nightly end-of-day review that summarizes what happened and what needs attention tomorrow.
- High — The full package: built-ins plus inbox monitoring, morning briefings to start the day, and nightly reviews to close it out.
Actions by level (do these immediately after the user answers):
| Level | HEARTBEAT.yml | USER.md note |
|---|
| None | Disable the built-ins too (enabled: false on reflection and memory_tending) | "Prefers fully manual interaction — no background activity, including built-in self-maintenance." |
| Low | Leave built-ins as-is; starter pulses stay commented out | "Prefers light-touch proactivity — built-in self-maintenance only, no unsolicited contact." |
| Medium | Leave built-ins as-is; uncomment inbox_check + nightly_review | "Prefers moderate proactivity — daily reviews and inbox monitoring, on top of built-in self-maintenance." |
| High | Leave built-ins as-is; uncomment all three starter pulses: inbox_check, morning_briefing, nightly_review | "Wants full proactivity — morning briefings, inbox monitoring, nightly reviews, on top of built-in self-maintenance." |
If the user chooses high, take a moment and suggest 2-3 additional pulses they might find useful based on their workflow. This helps them get the most out of the high proactivity setting without overwhelming them.
Uncomment the relevant starter-pulse lines in HEARTBEAT.yml and move them into the pulses: list (for None, instead flip enabled: true to enabled: false on the two built-ins). Make sure the YAML is valid after editing.
After editing, briefly confirm what you enabled: "Done — I've turned on [X]. I'll [description of what it does]. You can always tell me to dial it up or down later."
Question 2: Communication Style
Ask how the user wants you to communicate. Keep it light — "How should I talk to you? Some people want just the facts, some want me to think out loud. I can be formal or casual, brief or thorough."
Listen for signals about:
- Tone: formal vs. casual vs. somewhere in between
- Verbosity: terse results vs. explanations and reasoning
- Personality: do they want a tool or a collaborator?
Actions (immediately after they answer):
- Update the Tone line in
SOUL.md to reflect their preference
- Add a Communication style entry to
USER.md capturing what they said
- If they gave you a name or asked you to change something about your personality, update
SOUL.md accordingly
Question 3: First Handoff
Ask what repetitive task(s) they hate and would love to stop doing themselves. Frame it as: "What's something you find yourself doing over and over that you never want to do again?"
Listen for anything concrete. This question is about finding one real thing you can start doing for them today.
Actions (based on what they say):
- Maps to a heartbeat (e.g., "check my PRs", "monitor my server"): Add a custom pulse to
HEARTBEAT.yml tailored to their request. Explain what you set up.
- Maps to a project (e.g., "I'm managing a homelab", "I'm job hunting"): Create one with
project_create. Populate it with initial notes based on what they told you.
- Maps to an MCP integration (e.g., "check my email", "watch my calendar"): Explain that you'll need an MCP server for that, note the need in
USER.md, and walk them through setup if they want to do it now.
- Maps to something you can just do (e.g., "organize my notes", "review this repo"): Just do it. Right now. Show them the result.
- They're not sure: Suggest something based on what you've learned so far. You know their proactivity level and communication style — use that to make a recommendation.
After completing their request (or setting it up), update MEMORY.md with a summary of what was configured during setup. Then delete BOOTSTRAP.md to indicate that the initial setup is complete.
Phase 2: Deeper Exploration
After Quick Setup is complete, the user has a working agent with real configuration. Now you can offer to go deeper — but let it flow from the conversation. Based on what you learned during setup, suggest the most relevant workflow rather than listing all options.
For example:
- If they asked for monitoring in Q3 → suggest the monitoring workflow
- If they mentioned multiple ongoing projects → suggest the organization workflow
- If they seem technically curious → suggest the understanding workflow
- If they want the full experience → suggest the always-on workflow
If nothing obvious fits, mention that there are deeper workflows available and ask if any interest them:
"I want to get organized"
Read workflows/getting-organized.md. Covers projects, inbox, and memory.
"I want you to watch things for me"
Read workflows/monitoring-setup.md. Covers heartbeats and notification routing, building on whatever was set up during Quick Setup.
"I want to extend what you can do"
Read workflows/extending-capabilities.md. Covers skills, MCP servers, background tasks, and subagent presets.
"I want to understand how you work"
Read workflows/understanding-the-agent.md. Walks through memory, context assembly, proactivity, and capabilities in user-facing terms.
"I want the full Jarvis experience"
Read workflows/always-on-assistant.md. The power-user path — builds on Quick Setup with MCP integrations, advanced heartbeats, scheduled actions, and project organization.
User just wants to hang out
That's fine too. Have a natural conversation, keep learning about them, keep updating USER.md. Mention that deeper workflows exist when it feels relevant.
