Prep For

Prep For

Runtime skill — a playbook the deployed @context agent runs for its owner, invoked in natural language. Not a coding-agent workflow; those live in .agents/skills/.

Pull together a tight pre-meeting brief on a subject — a person, company, or project the owner is about to engage. Read-only: gather and synthesize, never file.

Procedure

1. Identify the subject. Pin down who/what from the request ("my 3pm with Sarah Lee from Acme" → person Sarah Lee, org Acme). If it's genuinely ambiguous and a wrong guess would waste the brief, ask one clarifying question instead of guessing.
2. Sweep what we already know — internal first. Run an entity sweep:
   • query_crm — contacts, notes, projects, and any reminders/meetings tagged to the subject (name, company, tags, meeting attendees). This is our relationship + history.
   • query_knowledge — knowledge-base prose about the subject (runbooks, summaries, "what I know about X").
   • query_slack (when connected) — recent threads mentioning the subject; the latest exchanges are often the freshest context in the brief.
3. Anchor to the meeting. Surface the specific upcoming meeting/reminder with the subject if there is one — the brief should serve that interaction. When the calendar source is connected, check query_calendar for the real event (time, attendees); when gmail is connected and there's an email thread with the subject, pull the latest exchange with query_gmail — the most recent thing they said is often the most useful line in the brief.
4. Widen to the web only for people/orgs we don't know. If the internal sweep turns up little or nothing on an external person or company (no contact on file, just a name), call query_web for public background — role, current company, recent news. Skip the web when we already have a solid internal picture, and for internal/private topics. Don't pad a brief with generic results, and keep the web query to public identity terms (name + company), not the owner's private notes about them.

Format

A short brief — only the sections that have content:

• Who — one line on the subject (role, company), tagged (on file) or (from web) so the owner knows the source.
• Our history — relationship, past touchpoints, what's open with them (from crm / knowledge / slack).
• This meeting — the anchoring meeting and the open items to raise.
• Background — public context, only when pulled from the web; mark it external / unverified.

Edge cases

• Total blank (nothing internal, nothing useful on the web): say so plainly and ask what context to add — don't fabricate a profile.
• Attribute every line to its source (crm, knowledge, slack, or web). Keep on-file facts and web background visibly separate; never blend an unverified web claim into our known record.