Meeting Prep

Prepare for upcoming meetings by gathering attendee context from multiple sources and producing a structured prep brief. Output follows ${CLAUDE_SKILL_DIR}/template.md. For detailed per-source lookup procedures, see ${CLAUDE_SKILL_DIR}/references/context-gathering.md.

---

Language

See METHODOLOGY.md language mirror rule. Reply in the user's language.

---

Step 1: Determine Meeting Selection

Smart routing based on user request:

If user names a specific meeting ("prep my 2pm", "prepare for the client meeting"):

• Proceed to Step 2 to fetch calendar events
• After fetching, match the user's description to a specific event (fuzzy match on time, title keywords, or attendee names)
• If no match found, ask for clarification

If user says "prep all my meetings" or just "prep me":

• Proceed to Step 2 to fetch all events in the next 24 hours
• Enter batch mode -- generate one brief per timed meeting

If user is vague or ambiguous:

• Fetch calendar events first (Step 2), then present options:

AskUserQuestion: "Which meeting would you like to prepare for?"
Options: [
  "[HH:MM] -- [Meeting Title] ([key attendees])",
  "[HH:MM] -- [Meeting Title] ([key attendees])",
  "Prepare for all meetings today",
  "I'll describe the meeting manually"
]

When building the options list:

• Separate all-day events from timed meetings. All-day events (birthdays, OOO, holidays) are NOT meetings to prep for -- surface them as context only (e.g., "Note: Alex is OOO today" if Alex is an attendee in another meeting)
• Flag tentative meetings the user has not responded to: "[HH:MM] -- [Title] (tentative -- you haven't responded)"
• Order timed meetings chronologically

---

Step 2: Fetch Calendar Events

Fetch events for the next 24 hours from Google Calendar.

Parameters:

• singleEvents=true -- expand recurring events into individual instances
• timeMin = current time (ISO 8601)
• timeMax = current time + 24 hours (ISO 8601)
• orderBy=startTime
• Fields requested: summary, start, end, attendees, description, status, htmlLink

Post-fetch filtering:

1. Remove events where status === "cancelled" -- these are deleted instances
2. Find the user's own entry in each event's attendees[] array (match by email from ~/Tandem/CLAUDE.md Tools & Connectors section, or ~/Tandem/memory/preferences.md)
3. Remove events where user's responseStatus === "declined" -- user won't attend
4. Separate all-day events (start.date field, no time component) from timed meetings (start.dateTime field) -- all-day events are context, not prep targets
5. Flag events where user's responseStatus === "needsAction" or "tentative": prompt user: "You haven't responded to [Title] at [Time]. Include in prep?"

If calendar is unavailable (no MCP connector, auth error, timeout):

• Jump to Step 2b (manual input fallback)
• For Claude Code CLI users, see ${CLAUDE_SKILL_DIR}/references/cli-fallback.md

---

Step 2b: Manual Input Fallback

When no calendar connector is available, or user chooses "describe the meeting manually":

AskUserQuestion: "Tell me about the meeting -- who's attending, what's the topic, and when is it?"
Options: [
  "I'll type the details now",
  "Let me paste the calendar invite"
]

Parse the user's response for:

• Attendee names (and companies/roles if mentioned)
• Meeting topic or title
• Time and date

Continue to Step 3 with whatever information was provided. Missing fields are OK -- the prep brief will use what's available.

---

Step 3: Gather Attendee Context

For detailed per-source lookup procedures, see ${CLAUDE_SKILL_DIR}/references/context-gathering.md.

Silently read from all 5 sources below. Do NOT show intermediate results, do NOT ask permission for each source. Compile everything, then present the final brief in Step 4.

Each source is independently optional. If a connector is unavailable or data is not found, skip that source silently -- never fail, never report gaps.

Source A: People Memory

For each attendee, read ~/Tandem/memory/people/{name}.md (lowercase, hyphens for spaces).

• Extract: role, company, relationship context, communication preferences, past interactions
• Read-only -- never create or modify people memory entries
• If file not found for an attendee, skip silently

Source B: Email Threads

Search Gmail for recent threads with each attendee:

• Query: from:{email} OR to:{email} newer_than:14d -category:promotions -category:updates
• Max 5 email subjects per attendee -- subjects only, NO full email bodies
• Summarize patterns: "3 recent threads with Alex about Q2 budget review"
• If Gmail is unavailable, skip silently

Source C: Kanban Tasks

Read ~/Tandem/tasks/kanban.md and search "To Do" and "In Progress" columns for tasks mentioning or assigned to each attendee.

• Max 3 tasks per attendee
• Note task status and any deadlines
• If kanban.md does not exist, skip silently

Source D: Monday.com Items

Search Monday.com for items assigned to or mentioning each attendee.

• Max 5 items per attendee
• Include board name, status, and due date if available
• If Monday.com MCP is unavailable, skip silently

Source E: Past Meeting Notes

Search for previous meetings with the same attendees:

• Glob ~/Tandem/meetings/* for files containing attendee names
• Also check ~/Tandem/documents/meeting-* for older meeting notes
• Pull key points, decisions, and unresolved items from the most recent 1-2 matches
• If no previous meeting notes exist, skip silently

Batch mode: For each meeting in the batch, run context gathering independently. Each meeting gets its own set of attendee lookups.

---

Step 4: Generate Prep Brief

Use the template from ${CLAUDE_SKILL_DIR}/template.md to structure the output.

Single Meeting Mode

Generate a full prep brief with:

• Executive summary (2-3 sentences): Who is in this meeting, what it is about, why it matters, and key context the user should know. This is the most important section -- if the user reads nothing else, this should prepare them.
• Attendee context grouped by relevance: Key attendees (most interaction history, external contacts, decision-makers) first, then others. Present as flowing narrative paragraphs, NOT per-person sections, NOT a table. Weave email threads, tasks, Monday.com items, and memory context together naturally.
• Open items relevant to meeting attendees or topic
• Previous meeting key points if available
• Talking points with exact words (see Ready-to-Use Scripts below)
• Footer listing sources that contributed data

If some attendees have no context available, simply do not mention them in the context section. No "gaps detected" or "no data found" notes anywhere.

Batch Mode

For each meeting, generate a compact brief (see template.md compact format). Lead with a batch overview containing a table: Time | Meeting | Attendees | Key Context.

Use compact format per meeting -- executive summary (1-2 sentences), key context (2-3 bullet points), and relevant open items (1-2 items or omit).

Saving the Brief

Filename convention: prep-YYYY-MM-DD-HHMM-slug.md

• Date and time from the event's start time in user's local timezone
• Slug: lowercase event title, spaces to hyphens, strip special characters, truncate to 40 chars
• Example: prep-2026-03-21-1400-client-review.md

Directory: ~/Tandem/meetings/

• Create the directory if it does not exist (first-time use)

Batch mode duplicate check: Before generating each brief:

1. Glob ~/Tandem/meetings/prep-{date}-{time}-*.md matching the event's date and time
2. If a matching file already exists, skip with message: "Skipping [Title] ([Time]) -- already prepped"

Save the brief using the Write tool.

---

Step 5: Log Stats and Sync

After saving the prep brief, log activity to ~/Tandem/stats.json.

Read ~/Tandem/stats.json. If it does not exist, create it as []. Parse the JSON array, append a new entry, and write it back:

{
  "type": "meeting-prep",
  "action": "created",
  "count": 1,
  "timeSavedMinutes": 15,
  "description": "Meeting prep: [Title] with [attendees] ([time])",
  "timestamp": "<current ISO 8601 UTC>"
}

For batch mode: log one entry per meeting prepped (not one aggregate entry).

After appending to stats.json, follow the /sync workflow from tandem-skills/core/sync/SKILL.md to rebuild ~/Tandem/dashboard.html with updated statistics. If /sync fails (e.g., dashboard.html does not exist yet), continue -- the main workflow must complete regardless of sync outcome.

---

Step 6: Confirm to User

After all briefs are saved and stats logged:

Single meeting mode:

• Show the executive summary inline
• Report: "Prep brief saved to ~/Tandem/meetings/[filename]"

Batch mode:

• List all meetings prepped with file paths:
  • "9:00 -- Team Standup -> prep-2026-03-21-0900-team-standup.md"
  • "14:00 -- Client Review -> prep-2026-03-21-1400-client-review.md"
• Note any skipped meetings (already prepped or declined)

No further prompts unless the user asks for changes or drill-down.

---

Ready-to-Use Scripts

Give the user exact words, not vague advice. For each meeting, include 2-3 ready-to-use scripts the user can say verbatim. Tailor scripts to the context gathered in Step 3.

Opening Lines

Provide 1-2 opening line options based on meeting context:

• Status update: "Let me give a quick update on [specific item] — we're at [status], and the main thing to flag is [key point]."
• Client meeting: "Great to connect again. Since our last conversation about [topic], here's where things stand..."
• First meeting with new contact: "Thanks for making the time. I've been looking forward to discussing [topic] — I'll start by sharing [context]."

Navigating Likely Questions

Based on open items and email threads, anticipate 2-3 likely questions and provide exact response scripts:

• If asked about timeline: "We're targeting [date] for [milestone]. The main dependency is [blocker]. If that clears by [date], we're on track."
• If asked about budget/cost: "The current estimate is [amount] for [scope]. I can break that down by [category] if that would be helpful."
• If asked about a problem: "Here's what happened: [brief fact]. Here's what we're doing about it: [action]. Expected resolution: [date]."

Closing / Next Steps

Provide a closing script:

• "To wrap up — the key action items are [1-2 items]. I'll [your action] by [date], and [attendee] will [their action]. Does that capture everything?"

Script Rules

• Scripts must use information from the prep context — never generic filler
• Match the meeting's formality level (casual for internal, professional for external)
• Include specific names, dates, and numbers from gathered context
• If insufficient context to write meaningful scripts, omit the section entirely rather than generating vague platitudes

---

Task Format Reference

Task format specification: tandem-skills/tasks/task-manager/references/kanban-format.md. When reading or moving tasks, preserve the existing format exactly.

---

Anti-Patterns

• Do NOT use the Bash tool anywhere -- all operations via allowed-tools and MCP
• Do NOT hardcode MCP tool function names -- describe the operation needed
• Do NOT show intermediate results during context gathering -- compile silently
• Do NOT include suggested talking points or agenda -- present context only (D-07)
• Do NOT include "gaps detected" or "missing data" sections -- skip silently (D-08)
• Do NOT write to people memory -- read-only access (D-11)
• Do NOT create per-person subsections in the brief -- group by relevance (D-06)
• Do NOT show empty sections with "None" placeholder -- omit entirely

---

Error Handling

• Calendar unavailable: Switch to manual input fallback (Step 2b). Never fail.
• All context sources empty: Still generate a brief with executive summary from calendar event details (title, attendees, description). The brief is useful even without enrichment.
• Gmail unavailable: Skip email context silently. Other sources may still provide value.
• Monday.com unavailable: Skip Monday.com context silently.
• Stats.json write fails: Log the error but do not block -- the prep brief is already saved and that is the primary deliverable.
• Sync fails: Continue -- dashboard update is secondary to prep brief delivery.