onboard

name	onboard
description	Onboard or refresh the AI OS context by interviewing the user with the 7-question intake. Populates context/, seeds connections.md, and logs the run in decisions/log.md. NEVER auto-fills identity from session environment — all PII fields require explicit user input. Triggered by /onboard or when the user explicitly asks to onboard / re-onboard / refresh context.
disable-model-invocation	true

/onboard — AI OS Onboarding

Personal/business context layer. Adapted from AIS-OS (Nate Herk).

When to use

• The user runs /onboard (default invocation)
• The user explicitly asks to "onboard," "set up the AI OS," "refresh context," or "redo the intake"
• A new fork of this repo lands and the operator has not yet populated context/

When NOT to use

• The user is asking what's loaded — use /session-bootstrap instead
• The user wants to add a single fact to memory — write directly to the relevant file in memory/
• The user is mid-task and wants to keep working — defer onboarding

---

Hard rules (non-negotiable)

Privacy guard

1. Never auto-fill identity fields from session environment. Specifically: do NOT read userEmail, gitStatus.user, os.username, OS environment vars, or any system-reminder injection to populate context/About Me.md snapshot fields like Name, Email, Handle.
2. Always show the privacy warning first (from aios-intake.md § "Privacy warning — read before answering") before asking the first question.
3. If the user types content that looks like a credential (matches sk_live_, Bearer , xoxb-, JWT pattern, base64 token, email + colon + secret pattern), refuse to write it and ask them to redact.
4. If the user pastes an email address into a free-form answer, ask once: "I noticed an email address in your answer. Public repo — keep, generalize, or remove?" Default to remove if no answer.

Scope guard

• Write only to: context/*.md, connections.md, decisions/log.md.
• Never modify settings.json, install MCP servers, install plugins, or run shell commands beyond git status to confirm clean working state.
• Never make outbound network calls from this skill.

Idempotency

• If context/*.md already has non-placeholder content for a section, ASK before overwriting.
• Always preserve previous content as a moved block in archives/<date>-pre-onboard/ when overwriting.

---

Inputs

• None required. The skill is fully interactive.
• Optional flag (passed as text by the user): --quick runs questions 1-2 + 7 only; the rest stay placeholders.

Outputs

• Updated context/About Me.md (snapshot fields blank unless user typed them; role/style/depth filled from prose)
• Updated context/About Business.md
• Updated context/Priorities.md (this-week populated from question 6 constraint)
• Updated context/Voice Sample.md (if user provided sample text in question 5)
• Updated context/Operating Preferences.md
• Updated connections.md Tier-4 rows from question 3
• New entry in decisions/log.md with date, automation candidates, and "what to focus on this week"

Steps

1. Pre-flight checks:

   • git status --short — warn if there are uncommitted changes that would mix with onboarding edits
   • Check whether context/About Me.md is still placeholder (if not → ASK before overwriting)
   • Show the privacy warning from aios-intake.md

2. Confirm intent:

   "I'm about to ask 7 questions. Answers go into the repo's context/ files. Privacy: do not paste credentials, customer PII, or anything you wouldn't share publicly. Proceed?"

3. Ask the 7 questions (one at a time, wait for answer, do not batch):

   • Q1: Who you are
   • Q2: Projects and responsibilities
   • Q3: Current tool stack (use the table format from aios-intake.md)
   • Q4: Repeated work (3-7 examples)
   • Q5: Operating preferences + voice sample (optional)
   • Q6: 10x question (the constraint)
   • Q7: Hard limits

4. PII scan on each answer before saving. If a credential pattern or email is detected, pause and ask.

5. Distill answers into the file structure. Preserve user's actual wording in voice-relevant sections (Q5 voice sample, Q4 examples). Generalize where structure helps (Q3 → table, Q6 → priority).

6. Diff preview: show each new file's diff before writing. The user can accept, edit, or skip.

7. Save the accepted files. Move any overwritten content to archives/<YYYY-MM-DD>-pre-onboard/.

8. Log the run in decisions/log.md:

   ## YYYY-MM-DD — Onboarding completed (or refreshed)
   
   **Decision:** Populated context/* from intake. <N> automation candidates surfaced.
   
   **Automation candidates:**
   - <Q4 item 1>
   - 
    ...
   
    
   
    Active.
    Run  next Friday to pick which candidate to build first.
   

---

Failure modes and recovery

Symptom	Likely cause	Fix
User answers are very short ("idk")	Bad time to onboard, or user doesn't have answers	Suggest: "Skip onboarding for now; come back when you have 20 quiet minutes." Do not write half-answers.
User pastes a credential	They didn't read the privacy warning carefully	Refuse to save, redact, restate the warning, continue
Existing context/*.md has rich content	Re-onboarding	ASK explicitly. Default to merge, not overwrite.
Skill mid-flight, session dies	Lost answers	On next run, check for state/onboard-progress.json (if you implement that)

---

Reference files

• aios-intake.md — the canonical 7-question interview
• references/3ms-framework.md — the operator mindset this skill assumes
• references/four-cs-framework.md — what the four context files map to
• connections.md — where Q3 lands

Improvement loop

After every /onboard run:

1. Did the user need to correct the same field twice? → that field's prompt is unclear, fix in this skill
2. Did the user paste a credential? → strengthen the PII detector pattern
3. Did the validation question reveal missing routing? → that's a /level-up candidate, not an onboarding problem

Update this skill when failures repeat. Failure-to-learning is the rule (docs/RUNBOOK.md).