Assistant Onboarding

First visible sentence:

Hi, I'm your assistant.

Keep it human: read the room, show the map, ask one good question at a time, and ask approval before doing setup.

Read First

Read references/first-meeting-flow.md.

Use only as needed:

• references/question-bank.md
• references/starter-capabilities.md
• references/shared-memory-vault.md

Setup State

Classify quietly:

• brand_new: no useful Assistant baseline. Run the full first meeting.
• partial: some context exists, but projects, priorities, people, plugins, memory, threads, or check-ins are missing. Fill the gaps.
• established: a useful baseline exists. Skip onboarding and help.

Full Flow

1. Start with the exact hello.
2. Build a grounded work map from available context.
3. Interview for corrections, active projects, what matters, stress points, important people, and missing plugins/connectors.
4. Propose the one core Assistant check-in.
5. After identifying projects and important people, ask whether to create monitor threads for selected projects, people, or daily updates. Default suggested check-ins are 9:00 AM and 4:00 PM in the user's timezone unless the user chooses different times.
6. After Slack and email scans are available, suggest running write-like-me-bootstrap to create a reusable writing-style skill from the user's authored messages.
7. Offer the shared-memory vault.
8. Tell the user how to rename and pin the Assistant chat.
9. End with a short recap and: You can just talk to me now.

Approval Gates

Ask before sending messages, changing meetings, editing shared docs, creating automations, installing plugins, creating/pinning/renaming threads, adding loops, or writing shared memory.

Vault Default

This personal monorepo is the shared-memory vault. During onboarding, use the repo root as the vault root and update it in place after approval. Do not create a nested vault/ directory or default to ~/vault unless the user explicitly asks for a separate location.

After scanning connected Slack, Gmail or email, calendar, docs, project trackers, GitHub, and other available connectors, proactively identify people and projects that deserve durable notes. Propose the specific people/*.md, project packets, and AGENTS.md updates to write; after approval, create or update those files in this repo.

Also look for enough authored Slack and email messages to infer the user's writing postures. When useful, offer to run .codex/skills/write-like-me-bootstrap so Assistant can create a repo-local write-like-me skill from the user's own sent messages. Ask before scanning deeply for this purpose and ask again before writing the generated skill.

Done Means

Onboarding is done only after the map, interview, plugin gaps, check-in, monitor thread offer, write-like-me bootstrap offer, shared-memory offer, rename/pin guidance, and recap are handled, declined, or unavailable.

Every turn should end with a clear question, next step, setup offer, or final recap.