📋 Meeting Prep Assistant

An entropy-aware meeting preparation skill that doesn't just aggregate data — it thinks about why the meeting exists and prepares you accordingly.

Most meeting prep tools treat every phase as data retrieval. They dump your calendar, paste in attendee names, and call it done. The result is a brief you could have written yourself in 30 seconds.

This skill is architecturally different. It was designed using Adaptive Narrative Control Theory (ANCT) — each phase uses the control mode matched to its actual uncertainty level:

• Data pull → DELEGATE (low entropy, just execute)
• Relevance judgment → NARRATE (medium entropy, requires interpretation)
• Meeting purpose inference → GENERATE → NARRATE (high entropy, requires exploring hypotheses then compressing)
• Brief output → adapts format to the meeting's entropy level

The insight: The highest-value phase — "what is this meeting really about?" — is the one most tools skip entirely.

🧠 Core Philosophy

• Prep is inference, not aggregation — pulling data is step 1 of 6, not the whole job
• Meetings have entropy levels — a recurring standup needs a bullet list, a skip-level needs deep framing
• Purpose is the product — the most valuable sentence in the brief is "what this meeting is really about"
• Format follows entropy — low-entropy meetings get quick briefs, high-entropy meetings get deep briefs
• Better to surface one insight than ten facts — synthesis over volume

🏗️ ANCT Architecture

This skill was designed with the adaptive-workflow-architect meta-skill. Here is its entropy profile and mode map:

Phase:      1          2          3          4          5            6
Entropy:    E1         E2         E3         E3         E4           E3→E2
Mode:       DELEGATE   DELEGATE   NARRATE    NARRATE    GEN→NAR      NAR→DEL

            pull       look up    select     prioritize  infer real   write
            metadata   attendees  relevant   open items  purpose      the brief
                                  notes                  (3→1)

Mode Transition Diagram

┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐
│ DELEGATE  │→│ DELEGATE  │→│ NARRATE   │→│ NARRATE   │
│ Pull      │  │ Lookup    │  │ Select    │  │ Prioritize│
│ metadata  │  │ attendees │  │ relevant  │  │ open      │
│           │  │           │  │ notes     │  │ items     │
│ E1        │  │ E2        │  │ E3        │  │ E3        │
└──────────┘  └──────────┘  └────┬─────┘  └────┬─────┘
                                  │              │
                 if empty → GENERATE    if empty → GENERATE
                            infer                  predict
                            context                questions

┌──────────────────────┐  ┌──────────────────────┐
│ GENERATE → NARRATE   │→│ NARRATE → DELEGATE    │
│ Assess real purpose   │  │ Write the brief       │
│ 3 hypotheses → pick 1 │  │ Format adapts to      │
│                       │  │ meeting entropy level  │
│ ⚡ COMPRESSION        │  │                       │
│ CHECKPOINT: 1-sentence│  │                       │
│ purpose required      │  │                       │
└──────────────────────┘  └──────────────────────┘

📥 Intake: Customize This Skill

{{MEETING_IDENTIFIER}}

{{CALENDAR_SOURCE}}

{{NOTE_SOURCES}}

{{TASK_SOURCES}}

{{RELATIONSHIP_CONTEXT}}

{{YOUR_ROLE}}

{{PREP_DEPTH}}

Phase 1: EXTRACT MEETING METADATA

Control Mode: DELEGATE | Entropy: E1 (Deterministic)

This is pure data pull. No judgment. Execute the pattern.

Actions

1. Pull from

   {{CALENDAR_SOURCE}}

   :
   • Meeting title
   • Date, time, duration
   • Attendee list (names + emails)
   • Agenda / description (if present)
   • Recurrence pattern (one-time, weekly, monthly, ad-hoc)
   • Meeting series history (how many past occurrences)
   • Location / video link

2. Classify meeting recurrence:

   • Recurring → there's history to pull (Phase 3 matters)
   • First-time → no history, Phase 5 (purpose inference) becomes critical
   • Ad-hoc → someone called this for a reason — Phase 5 is highest priority

Exit Condition

All metadata fields populated. If agenda is blank, flag for Phase 5 (purpose inference becomes more important).

Output

**Meeting:** [title]
**When:** [date, time, duration]
**Recurrence:** [pattern]
**Attendees:** [count] — [names]
**Agenda:** [text or "None provided"]
**Series history:** [N previous meetings]

Phase 2: IDENTIFY ATTENDEES + RELATIONSHIPS

Control Mode: DELEGATE | Entropy: E2 (Procedural)

Same lookup process every time, different people.

Actions

For each attendee, pull (from

{{RELATIONSHIP_CONTEXT}}

Escalation → NARRATE

If an attendee is unknown (no prior interaction, no context):

• Flag them in the brief
• In Phase 5, weight their presence as a signal about meeting purpose

Exit Condition

Every attendee has at minimum: name, inferred role, and last interaction date (or "unknown").

Output

**Attendees:**
- [Name] — [Title/Role] — Last interaction: [date] — [context note]
- [Name] — [Title/Role] — ⚠️ Unknown — first meeting with this person

Phase 3: PULL & SELECT RELEVANT NOTES

Control Mode: NARRATE | Entropy: E3 (Analytical)

This is where judgment begins. You're not pulling all notes — you're deciding which ones matter for this specific meeting.

Actions

1. Query

   {{NOTE_SOURCES}}

   for:
   • Previous meeting notes from same series (if recurring)
   • Notes mentioning attendees from Phase 2
   • Notes mentioning the meeting topic/agenda items
   • Recent notes (last 14 days) from related projects

2. Relevance filter (this is the NARRATE work):

   • Does this note contain decisions that are still open?
   • Does it reference commitments made by attendees?
   • Does it contain context the meeting will build on?
   • Is it recent enough to be actionable?

3. Select top 3-5 most relevant notes. For each, extract:

   • One-line summary
   • Key decisions or commitments
   • Unresolved items

Escalation → GENERATE (if notes are sparse)

If fewer than 2 relevant notes found:

"No substantial notes found. Based on the attendees ([names]) and topic ([agenda]), what context would likely exist? What has probably been discussed before?"

Generate inferred context rather than returning empty.

Exit Condition

3-5 relevant notes identified with key takeaways, OR generated context if notes don't exist.

Output

**Relevant Context:**
1. [Date] — [Note summary] — Key point: [decision/commitment/open item]
2. [Date] — [Note summary] — Key point: [decision/commitment/open item]
3. [Inferred] — Based on attendees and topic, likely prior context: [inference]

Phase 4: SURFACE OPEN ITEMS

Control Mode: NARRATE | Entropy: E3 (Analytical)

Not just "what's open" but "what's open that matters for THIS meeting."

Actions

1. Query

   {{TASK_SOURCES}}

   for:
   • Items assigned to you related to this meeting's topic
   • Items assigned to attendees that you're involved in
   • Items mentioned in Phase 3 notes that are still open
   • Overdue items involving any attendee

2. Prioritization (NARRATE work):

   • Which items are the attendees likely to ask about?
   • Which items have been open long enough to create tension?
   • Which items block something the meeting will discuss?

3. Rank by relevance to this meeting. Top 3-7 items.

Escalation → GENERATE (if no open items found)

If task systems return nothing relevant:

"No tracked items found. Given the meeting topic and attendees, what questions should you be prepared to answer? What might someone ask you for a status on?"

Generate anticipated questions rather than returning empty.

Exit Condition

Prioritized list of 3-7 open items, OR generated anticipated questions.

Output

**Open Items:**
- 🔴 [Overdue/urgent] [item] — owner: [who] — status: [status]
- 🟡 [Active] [item] — owner: [who] — likely to come up because [reason]
- 🟢 [FYI] [item] — context for discussion

**If nothing tracked:**
- ❓ Be ready to discuss: [anticipated question]
- ❓ Possible status request: [topic]

Phase 5: ASSESS REAL MEETING PURPOSE

Control Mode: GENERATE → NARRATE | Entropy: E4 (Creative)

This is the highest-entropy phase and the most valuable.

The agenda says "sync" or "status update" or "discuss Q3 plan." That's the stated purpose. The real purpose is often different:

• A "status update" might actually be "I need to know if I can trust this timeline"
• A "1:1" might actually be "I'm about to give you feedback"
• A "brainstorm" might actually be "I already have an answer, I want buy-in"

GENERATE Step (expand)

Using all context from Phases 1-4, generate 3 hypotheses about why this meeting really exists:

Hypothesis 1 (surface): [What the agenda says]
Hypothesis 2 (deeper):  [What the attendee mix suggests]
Hypothesis 3 (deepest): [What the timing, recurrence, and open items suggest]

Signals to weigh:

• Who called the meeting? Their role + seniority = intent signal
• Who's invited who isn't normally there? New attendees = new agenda
• What just happened? Recent events that would trigger this meeting
• What's the cadence delta? A weekly that got moved to daily = urgency signal
• Is there an agenda? No agenda on a meeting with senior people = the topic is sensitive

⚡ COMPRESSION CHECKPOINT

You cannot proceed to Phase 6 until the 3 hypotheses are compressed to a single sentence:

"This meeting is really about: _______________"

NARRATE Step (compress)

Select the most likely purpose. State it in one sentence. Note the runner-up hypothesis as an alternative reading.

Exit Condition

One-sentence primary purpose + one-sentence alternative reading.

Output

**What this meeting is really about:**
[One sentence — primary hypothesis]

**Alternative reading:**
[One sentence — second hypothesis]

Phase 6: GENERATE THE PREP BRIEF

Control Mode: NARRATE → DELEGATE | Entropy: E3 → E2

Synthesize all phases into a prep brief. Format adapts to meeting entropy.

Meeting Entropy Classification

Brief Format A: Quick Prep (Low-Entropy Meeting)

For recurring meetings with familiar attendees and clear agendas.

{{PREP_DEPTH}}

## [Meeting Title] — Quick Prep
**[Date] [Time] ([Duration])**

**Purpose:** [one sentence from Phase 5]

**Attendees:** [names + roles, one line]

**Your items:**
- [open item 1]
- [open item 2]

**Since last time:**
- [key change from recent notes]
- [key change from recent notes]

**One thing to know going in:**
[single most important context point]

Brief Format B: Standard Prep (Medium-Entropy Meeting)

For meetings with some uncertainty — mixed attendees, partially clear agenda.

{{PREP_DEPTH}}

## [Meeting Title] — Prep Brief
**[Date] [Time] ([Duration])**

**What this meeting is really about:**
[one sentence from Phase 5]

**Attendees:**
- [Name] — [Role] — [one-line context]
- [Name] — [Role] — [one-line context]

**Relevant context:**
- [Key point from notes 1]
- [Key point from notes 2]

**Open items likely to surface:**
- 🔴 [urgent item]
- 🟡 [active item]

**Your position going in:**
[What do you think/want/need from this meeting?]

Brief Format C: Deep Prep (High-Entropy Meeting)

For first-time meetings, skip-level conversations, strategy sessions, meetings with vague agendas and senior attendees.

{{PREP_DEPTH}}

## [Meeting Title] — Deep Prep
**[Date] [Time] ([Duration])**

**What this meeting is really about:**
[one sentence — primary hypothesis from Phase 5]

**Alternative reading:**
[one sentence — what else this could be about]

**Key relationships in the room:**
- [Name] — [Role] — [dynamic: ally, new stakeholder, decision-maker, observer]
- [Name] — ⚠️ Unknown — [what their presence might signal]

**Context you need:**
- [Synthesized point from notes — not a raw note, a takeaway]
- [Synthesized point from notes]
- [Inferred context if no notes exist]

**What you should be ready for:**
- [Anticipated question or topic 1]
- [Anticipated question or topic 2]
- [Uncomfortable possibility to be aware of]

**Open threads that may surface:**
- 🔴 [item + whose it is + why it's tense]
- 🟡 [item + status + what's expected of you]

**Your position:**
[What do you think going in? What do you want out of this meeting?]
[What would success look like when it ends?]

**One sentence to have ready:**
[If someone puts you on the spot, this is your grounding statement]

⚠️ Anti-Patterns

MEETING PREP ANTI-PATTERNS (from ANCT failure mode analysis):

✗ "Data dump" brief
  Lists attendees, pastes raw notes, shows all open items.
  No synthesis. No purpose inference.
  → ANCT diagnosis: all-DELEGATE, skipped NARRATE and GENERATE phases

✗ "Every meeting gets the same brief"
  Recurring standup gets the same format as a board prep.
  → ANCT diagnosis: flat entropy assumption, no format adaptation

✗ "First hypothesis accepted"
  "It says 'status update' so it's a status update."
  → ANCT diagnosis: premature compression, skipped GENERATE in Phase 5

✗ "No brief because no notes"
  Notes don't exist, so the skill returns nothing.
  → ANCT diagnosis: missing escalation, should GENERATE inferred context

✗ "20-item open item list"
  Everything remotely related is included. No prioritization.
  → ANCT diagnosis: DELEGATE applied to E3 phase, no NARRATE filter

CORRECT PATTERNS:

✓ Phase 5 always runs (even for "obvious" meetings)
✓ Brief format matches meeting entropy level
✓ Empty data triggers generation, not silence
✓ Open items filtered to "what THIS meeting will care about"
✓ One sentence of purpose is worth more than ten facts

📚 Example Applications

🌐 Platform Notes

/skill meeting-prep-assistant

Compliance

• PII Risk: Medium. Meeting attendees, calendar data, and notes may contain names, emails, and organizational context. All processing is session-local. No data is stored by the skill itself. Users should not paste sensitive meeting content into public LLM interfaces.
• Model Recommendation: Sonnet for standard prep (Phase 5 inference benefits from strong reasoning). Haiku for quick prep on low-entropy recurring meetings.
• Human Oversight: The brief is a proposal. The user reads it, adjusts their framing, and walks into the meeting with their own judgment — not the agent's.

🔗 Integration: Inbox Intelligence Compendium

This skill becomes significantly more powerful when paired with the inbox-intelligence skill's knowledge compendium.

How They Connect

The inbox-intelligence skill builds a persistent, cross-referenced knowledge base from your email and Teams messages. When meeting-prep-assistant can query that compendium, several phases get richer inputs:

How to Use Together

1. Run

   inbox-intelligence

   on your inbox regularly (daily or before prep sessions)
2. The compendium builds at

   {{COMPENDIUM_PATH}}

3. When running

   meeting-prep-assistant

   , set

   {{NOTE_SOURCES}}

   to include

   compendium

4. Phases 2-5 automatically query the compendium for richer context

What This Enables

• "Why was this meeting called?" — The compendium knows what email threads preceded the calendar invite
• "What will they ask me about?" — Cross-reference attendee communication patterns with open topics
• "What happened since our last meeting?" — Compendium timeline shows every relevant interaction, not just formal notes
• Prediction accuracy — Phase 5 purpose inference improves dramatically when the agent can see the communication trail that led to the meeting

Design Credit

This skill's architecture was designed using the adaptive-workflow-architect meta-skill, applying Adaptive Narrative Control Theory (ANCT) to map each phase to its optimal control mode based on entropy level.