Meeting Prep

Gathers cross-channel context before any meeting so you walk in fully prepared. Works for person-based prep (1:1s) and topic-based prep (product syncs, reviews).

When to Activate

• User says "meeting prep", "prep for my meeting with X", "prep for 1:1"
• User provides a person name or meeting topic to prepare for
• User says "what should I discuss with X?"

Input

The user provides ONE of:

• Person name: e.g., "Sarah Chen", "Marco", "the CTO"
• Topic/meeting name: e.g., "Q3 planning", "API migration review"
• Calendar event: a specific meeting from today's calendar

If no input, check Google Calendar MCP for the next upcoming meeting and prep for that.

Execution

Context Check

Before gathering, read

context/company.md

context/personas.md

Step 0: Resolve Target

• Person name given: use as search target across all sources.
• Topic given: use as search query.
• No input: fetch next meeting from Google Calendar, extract attendees and title.

Step 0.5: Org-Survival + People File Check (Person-Based Only)

Before gathering from external sources, read two local files:

1. memory/org-survival.md

   — Check if person has an entry. Extract: what they want, risks they carry, recommended approach. Surface as Political Context block (2–3 lines max). Skip silently if not found.

2. memory/people/{name}.md

   — Check for accumulated meeting history with this person. Extract: prior commitments they made (were they fulfilled?), communication style, known preferences. Fold into Their Current Focus and Suggested Talking Points.

This is the highest-signal step — it's the context no live API can replicate.

Step 1: Parallel Data Gathering (Fan-Out)

Launch parallel searches across all connected MCP sources.

For person-based prep:

assignee = "{person}" OR reporter = "{person}" AND updated >= -14d

contributor = "{person}" AND lastModified > now("-14d")

For topic-based prep:

text ~ "{topic}" AND updated >= -30d

text ~ "{topic}" AND lastModified > now("-30d")

Step 2: Synthesize Prep Doc

Merge all results into the output format below. Deduplicate across sources. Prioritize recent items.

Output Format

Output directly in conversation (do NOT create a file unless asked):

# Meeting Prep: {Person or Topic}
**For:** {Meeting name if known} | **When:** {Time if known}

## Political Context _(person-based, if found in org-survival.md)_
- {What they want from this interaction}
- {Any risk or tension to navigate}
- {Recommended framing or approach}

## Last 3 Interactions
1. **{Date}** ({source}): {1-2 sentence summary}
2. **{Date}** ({source}): {1-2 sentence summary}
3. **{Date}** ({source}): {1-2 sentence summary}

## They're Waiting On You
- {Action item you owe them, with source}
(If none found: "Nothing outstanding found.")

## You're Waiting On Them
- {Action item they owe you, with source}
(If none found: "Nothing outstanding found.")

## Their Current Focus
- {What they're working on based on Jira/GitHub/Slack}

## Open Threads
- {Unresolved Slack threads or email chains}
- {Blocked Jira issues}

## Suggested Talking Points
1. {Based on open items and recent activity}
2. {Based on unresolved threads}
3. {Based on upcoming deadlines or decisions}

Rules

• Recency bias: Prefer last 14 days for people, last 30 days for topics.
• No speculation: Only include items found in actual sources. Skip empty sections.
• Cite sources: Always note where information came from.
• Keep it scannable: Readable in 2 minutes before a meeting.
• Respect privacy: Only include conversations where the user is a participant.

Fallback

If MCP tools are unavailable:

1. Note which sources were unreachable.
2. Produce the prep doc with whatever was found.
3. Suggest the user check those sources manually.