capturing-meeting-intelligence

Connects Otter.ai (transcription), Granola (structured meeting notes), and Mem (knowledge storage) into a unified meeting intelligence workflow for Loxihart and FlashFusion. Routes each meeting to the

SKILL.md

GitHub
<!-- ─────────────────────────────────────────────────────────────────────── -->

DEPENDENCY_DECLARATION

Tool MCP Status Required / Optional Role in Workflow
Otter.ai ✅ Connected Required — primary transcript source for long/recorded meetings (30+ min) Live transcription, highlight capture, searchable audio archive
Granola ✅ Connected Required — primary structured notes source for 1-on-1s and short meetings AI-structured note generation, automatic meeting summary
Mem ✅ Connected Required — distilled knowledge storage and retrieval Semantic PKM storage; destination for three-item extract
Notion ✅ Connected Required — project and client record routing destination Structured record-keeping per meeting type (FlashFusion clients, Loxihart, workshops)
<!-- ─────────────────────────────────────────────────────────────────────── -->

Meeting Intelligence Capture

Three tools, one workflow. Raw capture lives in Otter or Granola. Distilled context lives in Mem. Nothing useful lives only in your head.


Evals

Should trigger:

  1. "I just got off a call with a FlashFusion client — help me process the notes"
  2. "Pull up what we decided in the last workshop I ran at Loxihart"
  3. "Start Otter for this — it's going to be a long one"

Should NOT trigger:

  1. "Write a meeting agenda" — pre-meeting prep, not capture
  2. "Draft a follow-up email from these notes" — use message_compose or drafting skill

Decision Tree: Which Tool to Use

What kind of meeting is this?
  │
  ├─ Workshop / training / recorded call / 30+ minutes
  │    → Otter.ai (full transcript, searchable, AI summary)
  │
  ├─ 1-on-1 / project call / structured agenda / under 30 minutes
  │    → Granola (AI-structured notes, less noise than full transcript)
  │
  ├─ Quick sync / hallway conversation / ad hoc decision
  │    → Mem directly (capture the decision or context only, no transcript)
  │
  └─ Already have transcript or notes from another tool
       → Skip to Post-Processing (Section 3)

Full routing rationale in references/tool-routing.md.


Section 1: Before the Meeting (2 minutes)

  1. Determine tool (decision tree above)
  2. Start the tool before the meeting begins — not after
  3. Set a Mem reminder or calendar note: "Process [meeting name] by EOD"
  4. If FlashFusion client call: open the client's Notion record in another tab

Section 2: During the Meeting

Otter.ai: let it run. Add highlights (⭐ button) at: decisions made, action items committed to, important context shared by the other party. You don't need to take separate notes — highlights are enough.

Granola: let it run in the background. Granola structures notes automatically. Add a quick typed note for anything the AI is likely to miss: proper nouns, unusual terms, decisions stated informally.

Mem: if the meeting is a quick sync, open a Mem note immediately and type the key point(s) as they happen. One sentence per decision or action.


Section 3: Post-Processing Protocol

Run within 24 hours. Context degrades fast. A transcript you process 24 hours later is twice as useful as one you process a week later.

Step 1: Review the capture

  • Otter: scan the AI summary + your highlights. Are the highlights complete? Did Otter miss any key decisions?
  • Granola: read the structured notes. Add any context Granola structured incorrectly or missed.

Step 2: Extract the three things that matter

From any meeting, extract exactly three things:

DECISIONS: [What was decided — specific, not "discussed X"]
ACTIONS: [Who does what by when — named, not "someone will"]
CONTEXT: [Background knowledge that makes future interactions smarter]

If you can't identify at least one item in each category, the meeting may not have needed to happen. Note that for your own future scheduling.

Step 3: Store in Mem

Create or update a Mem note. Title format: {{Meeting type}} — {{Other party}} — {{YYYY-MM-DD}}

Paste the three-item extract. Add any relevant links (ticket, Notion page, SOW).

Do not paste the full transcript into Mem. The transcript lives in Otter/Granola. Mem gets the distilled version — what you'd tell someone if they asked "what came out of that meeting?"

Step 4: Route to the right Notion record

Meeting type Notion destination
FlashFusion client call FlashFusion/Clients/{{Name}}/Calls/
Loxihart internal meeting Loxihart notes space
AI coaching conversation Append to AI adoption coaching log
Workshop (Loxihart or FlashFusion) Session record in workshop folder
Strategic/personal Personal knowledge base

Section 4: Retrieval

When you need to recall something from a past meeting:

  1. Mem first — search by person name, topic, or date. Mem's semantic search handles "what did we decide about the FlashFusion pricing call" better than keyword search.

  2. Otter second — if you need the verbatim quote or the exact moment something was said, search Otter by keyword.

  3. Granola third — for structured notes from 1-on-1s, Granola's organization by meeting makes chronological review fast.

The Mem note should make Otter/Granola lookups rare. If you're frequently going back to the raw transcript, the distillation step isn't being done.


Self-Verification

After any post-processing session:

  • Three-item extract (decisions, actions, context) is in Mem
  • Mem note is linked to the relevant Notion record
  • Action items are in whatever task system you use (not just in the note)
  • Otter/Granola transcript is accessible but not the primary reference

<!-- ─────────────────────────────────────────────────────────────────────── -->

FAIL_IF CONDITIONS

Otter.ai unavailable:

  • FAIL_IF Otter.ai MCP returns auth error, connection failure, or recording cannot be started: immediately switch to Granola as the primary capture source regardless of meeting length or type; do not wait or retry Otter during an active meeting.
  • DEGRADATION PATH: Granola handles recording and structure. Post-processing follows the same three-item extract protocol using Granola output. Note in the Mem entry: "Otter unavailable — full audio transcript not available for [date]; Granola structured notes are the primary source."

Granola unavailable:

  • FAIL_IF Granola MCP returns error or structured notes cannot be retrieved after a meeting: use the Otter.ai transcript directly as the primary post-processing source; skip the structured-notes review step; proceed with raw transcript scan to extract decisions, actions, and context.
  • DEGRADATION PATH: do not delay the 24-hour post-processing window waiting for Granola to recover. Extract the three-item summary from Otter directly and store in Mem. Note in the Mem entry: "Granola unavailable — distilled from raw Otter transcript; structured notes may be incomplete."

Mem unavailable:

  • FAIL_IF Mem MCP returns auth error or write failure during the post-processing step: route the three-item extract directly to the relevant Notion record as a fallback storage destination; prepend the note with "⚠️ PENDING MEM SYNC — Mem unavailable on [date]; logged to Notion temporarily."
  • DEGRADATION PATH: Notion is the temporary PKM destination. When Mem is restored, sync the Notion-logged entries to Mem, update the Mem title and tags as normal, and remove the PENDING MEM SYNC flag from the Notion entry. Note the affected date range in a Mem entry titled "PKM GAP — [date range]" so future retrieval searches know to check Notion for that window.

Notion unavailable:

  • FAIL_IF Notion MCP returns connection error or page write failure during the routing step: do not block the Mem distillation step; the Mem note is the primary artifact and stands complete without Notion routing.
  • DEGRADATION PATH: log the intended Notion routing destination as a tag directly in the Mem note body (e.g., [[NOTION-PENDING: FlashFusion/Clients/{{Name}}/Calls/ — route when restored]]). Create a follow-up task in your task system to complete Notion routing when the connection is restored. The meeting intelligence is fully captured; Notion routing is deferred, not skipped permanently.
<!-- ─────────────────────────────────────────────────────────────────────── -->

CLAIMS

  • Processing capture within 24 hours produces meaningfully better distilled notes than waiting; the meeting's context is still active in working memory and gaps are fillable
  • Distilled Mem notes retrieve faster and more accurately than searching raw transcripts for the same information

COUNTEREXAMPLE

  • For legal or compliance-relevant conversations (client disputes, HR matters), the raw transcript is the primary artifact — do not rely on distillation alone; keep the full Otter recording

CONTRADICTIONS

  • "Use Granola for short meetings" conflicts with situations where a 30-minute meeting is densely technical; use Otter for technical density regardless of length