Limitless Overnight Analysis

Deep analysis of today's Limitless lifelog recordings. The user wears a Limitless AI pendant that captures all conversations throughout the day -- both scheduled meetings and impromptu conversations. This skill reviews everything that was captured and surfaces what matters.

Why This Exists

Meetings captured by Fireflies have structured transcripts. But a lot of important information flows through conversations that aren't formal meetings -- phone calls, hallway chats, thinking-out-loud moments. Limitless captures all of it. This skill reviews the full day and extracts what Edwin needs to know.

Data Gathering

1. Read today's Limitless lifelogs

Read all files matching: data/limitless/lifelogs/YYYY-MM/YYYY-MM-DD-*.md

Each file is a conversation segment with:

• Time range and duration
• Summary
• Speaker-attributed transcript

2. Cross-reference with calendar

Read today's calendar (data/o365/calendar/YYYY-MM/YYYY-MM-DD.md and data/google/calendar/YYYY-MM/YYYY-MM-DD.md).

For each Limitless segment, determine:

• Was this a scheduled meeting? (match by time overlap with calendar events)
• Was this an off-calendar conversation? (no matching event = valuable ambient capture)
• Was this a phone call? (check data/calls/ for matching timeframe)

3. Read today's Fireflies transcripts

Read data/fireflies/transcripts/YYYY-MM/YYYY-MM-DD-*.md.

Identify which Limitless segments are already covered by Fireflies (better quality transcripts). Focus Limitless analysis on segments NOT covered by Fireflies.

Analysis

For each Limitless segment that isn't already covered by Fireflies:

Conversation Classification

• Work meeting (captured by Limitless but not Fireflies)
• Phone call (personal or work)
• Ambient/hallway conversation
• Solo thinking (talking to oneself or dictating)
• Family conversation (keep private -- extract only logistics that affect schedule)
• Other

Content Extraction

For each non-Fireflies segment:

1. Key points -- what was discussed, any decisions made
2. Commitments -- anything that should be a PM item (flag for pm-capture to pick up, or add directly)
3. Emotional signals -- the user's energy level, frustration points, excitement about something
4. People mentioned -- anyone referenced who might need follow-up
5. Context for tomorrow -- anything that should inform tomorrow's morning brief

Pattern Analysis

Across all segments:

1. Time allocation -- how many hours in meetings vs. unstructured time vs. heads-down work?
2. Who did the user talk to most today? Cross-reference with org chart / priority relationships.
3. Energy arc -- how did energy/mood change through the day? (early morning vs. afternoon vs. evening)
4. Unaddressed items -- things that came up in conversation but didn't get resolved or captured anywhere

Output

Write analysis to: ~/Edwin/briefing-book/docs/5. Overnight/logs/YYYY-MM-DD-limitless.md

Format:

---
date: YYYY-MM-DD
type: limitless-analysis
segments: [count]
off_calendar: [count]
---

# Limitless Day Review -- YYYY-MM-DD

## Day Overview
[2-3 sentences: what kind of day was it? Heavy meetings? Lots of phone calls? Quiet heads-down?]

## Off-Calendar Conversations
[These are the gold -- conversations Fireflies didn't capture]

### [Time] -- [Brief description]
- **Who:** [participants if identifiable]
- **Key points:** [what was discussed]
- **Action items:** [if any]
- **Context:** [why this matters]

## Emotional Arc
[How was the day from an energy/stress perspective? What moments were high/low?]

## People Map
[Who did the user interact with today and in what context?]

## Tomorrow Context
[What from today should inform tomorrow's brief or Edwin's behavior?]

## PM Items Extracted
[List any new PM items added, or note if none]

After writing, publish to Obsidian:

python3 briefing-book/scripts/obsidian-publish --all

Completion Report

SKILL_COMPLETE: limitless-analysis
STATUS: success | error
ARTIFACT: ~/Edwin/briefing-book/docs/5. Overnight/logs/YYYY-MM-DD-limitless.md
SEGMENTS_REVIEWED: [count]
OFF_CALENDAR: [count]
PM_ITEMS_EXTRACTED: [count]
ERRORS: [any errors, or "none"]