meeting-brief

# Meeting Brief

Automated daily meeting preparation system that researches meeting attendees and sends you personalized briefs.

## What It Does

Every morning (configurable time):
1. Checks your calendar for today's meetings (via gcalcli)
2. Extracts attendees from each meeting
3. Filters out your team members (configurable)
4. Deep researches each external person:
   - LinkedIn profile (web search)
   - Company information
   - GitHub profile (if engineer)
   - Past interactions/notes (memory search)
   - Recent news/activity
5. Generates AI-powered brief per person
6. Sends 1 email per person to your inbox

## Setup

### 1. Configure Team Members

Edit `config.json` to list your team members (these will be skipped):

```json
{
  "team_members": [
    "alice@yourcompany.com",
    "bob@yourcompany.com",
    "team@yourcompany.com"
  ],
  "team_domains": [
    "@yourcompany.com"
  ],
  "schedule": "0 7 * * *",
  "timezone": "America/Los_Angeles",
  "your_email": "you@yourcompany.com",
  "brief_from": "Meeting Brief <briefbot@yourcompany.com>",
  "slack_webhook": "https://hooks.slack.com/services/YOUR/WEBHOOK/URL",
  "send_email": true,
  "send_slack": true,
  "include_calendar_details": true,
  "research_depth": "standard"
}
```

**Config options:**
- `team_members`: Emails to skip (exact match)
- `team_domains`: Domain patterns to skip (e.g., skip all @yourcompany.com)
- `schedule`: Cron expression for daily run (default: 7am)
- `timezone`: Timezone for schedule
- `your_email`: Where to send briefs
- `brief_from`: From address for briefs
- `slack_webhook`: Slack incoming webhook URL (optional)
- `send_email`: Whether to send email briefs (default: true)
- `send_slack`: Whether to send Slack notifications (default: false)
- `include_calendar_details`: Include meeting time/location in brief
- `research_depth`: `quick` (web only), `standard` (web + GitHub), `deep` (web + GitHub + past notes)

### 2. Set Up Daily Cron Job

Use OpenClaw's cron tool to schedule the daily run:

```javascript
// Create cron job for daily meeting briefs
{
  name: "Meeting Brief - Daily",
  schedule: {
    kind: "cron",
    expr: "0 7 * * *",  // 7 AM daily (UTC)
    tz: "America/Los_Angeles"
  },
  payload: {
    kind: "agentTurn",
    message: "Run the meeting-brief skill for today's meetings",
    timeoutSeconds: 600
  },
  sessionTarget: "isolated"
}
```

Alternatively, run manually:
```bash
cd skills/meeting-brief
./scripts/run_daily.sh
```

## How It Works

### Main Workflow (`scripts/run_daily.sh`)

1. **Fetch today's meetings** (`scripts/check_calendar.sh`)
   - Uses gcalcli to get today's agenda
   - Parses meeting times, titles, attendees
   - Outputs JSON with meeting details

2. **Filter external attendees** (built into run_daily.sh)
   - Loads config.json
   - Filters out team members and team domains
   - Creates list of people to research

3. **Research each person** (`scripts/research_person.js`)
   - Web search: LinkedIn profile, company info
   - GitHub search: Profile and repos (if applicable)
   - Memory search: Past interactions/notes
   - News search: Recent activity
   - Outputs structured research JSON

4. **Generate brief** (`scripts/generate_brief.js`)
   - Uses OpenClaw session to generate AI brief
   - Inputs: research data + meeting context
   - Outputs: Two formats:
     - **Email**: Concise bullet-point brief
     - **Slack**: Rich paragraph-style story with deeper context and narrative

5. **Send brief** 
   - Email: Uses Gmail skill (`send_email: true`)
   - Slack: Uses webhook (`send_slack: true`, `scripts/send_slack.sh`)
   - Subject: "Meeting Brief: [Person Name] - [Meeting Title]"
   - Body: AI-generated brief with research

6. **Save to personal CRM** (`supernotes/people/`)
   - Each researched person saved as markdown file
   - Includes: research data, meeting context, date
   - Builds personal relationship database over time

7. **Track sent briefs** (logs to `data/sent/YYYY-MM-DD.json`)
   - Prevents duplicates
   - Enables analytics

## Research Process

For each external attendee, the system researches:

### Web Search (Always)
- LinkedIn profile (name + company)
- Company information
- Recent news mentions
- Professional background

### GitHub (If `research_depth` is `standard` or `deep`)
- GitHub profile lookup (by name/email)
- Recent repos and contributions
- Technical focus areas

### Memory/Past Notes (If `research_depth` is `deep`)
- Search MEMORY.md and daily notes
- Past meeting notes
- Previous interactions
- Context from past conversations

### Output Format

Research is structured as JSON:

```json
{
  "person": {
    "name": "Jane Doe",
    "email": "jane@example.com",
    "company": "Example Corp",
    "title": "VP Engineering"
  },
  "linkedin": {
    "url": "...",
    "bio": "...",
    "experience": [...]
  },
  "github": {
    "username": "janedoe",
    "profile_url": "...",
    "recent_repos": [...]
  },
  "company": {
    "name": "Example Corp",
    "industry": "...",
    "recent_news": [...]
  },
  "past_interactions": [
    "Met at conference in 2024",
    "Discussed partnership opportunity"
  ]
}
```

## Brief Generation

The AI-generated brief comes in **two formats**:

### Email Format (Concise Bullets)

1. **Quick Overview**
   - Who they are (name, title, company)
   - Why you're meeting (meeting title/description)

2. **Background**
   - Professional background (LinkedIn)
   - Company context
   - Technical expertise (GitHub, if applicable)

3. **Conversation Starters**
   - Based on recent activity
   - Shared interests/connections
   - Relevant topics

4. **Action Items / Notes**
   - Past interactions (if any)
   - Things to remember
   - Follow-up items

### Slack Format (Rich Story)

Deeper, narrative-driven brief with:
- Paragraph-style storytelling about the person
- Context about their journey and recent work
- Compelling hooks and conversation angles
- More background color and detail
- Stronger narrative flow than bullet points

**Example Brief:**

```
Subject: Meeting Brief: Jane Doe - Product Partnership Discussion

Hi,

You're meeting with Jane Doe today at 2pm.

## Quick Overview
Jane is VP of Engineering at Example Corp, a B2B SaaS company in the dev tools space. She's been there for 3 years and previously worked at GitHub and Microsoft.

## Background
- Strong background in developer tooling and infrastructure
- Recently led Example Corp's API platform overhaul (launched Q4 2025)
- Active on GitHub (janedoe) - maintains several open-source CLI tools
- Technical blog focuses on API design and developer experience

## Conversation Starters
- Their new API platform (just launched, getting good traction)
- Recent blog post on GraphQL vs REST (published last week)
- Shared interest in developer experience (noted in her LinkedIn)

## Notes
- You met briefly at DevTools Summit 2024
- She mentioned interest in partnering on integration opportunities

---
Meeting: Product Partnership Discussion
Time: Today at 2:00 PM
Location: Zoom (link in calendar)
```

## Manual Usage

Run for a specific person:

```bash
# Research a person
node scripts/research_person.js "Jane Doe" "jane@example.com" "Example Corp"

# Generate brief
node scripts/generate_brief.js research_output.json meeting_context.json

# Send brief
./scripts/send_brief.sh brief.html "Jane Doe"
```

Run for today's meetings:

```bash
./scripts/run_daily.sh
```

## Data & Logs

```
meeting-brief/
├── data/
│   ├── sent/              # Sent brief logs (by date)
│   │   └── 2026-02-21.json
│   ├── research/          # Research cache (by person)
│   │   └── jane-doe.json
│   └── meetings/          # Meeting data (by date)
│       └── 2026-02-21.json
└── logs/
    └── run.log            # Execution logs
```

## Tips

1. **Test with dry-run first**: Set `DRY_RUN=true` in run_daily.sh to preview without sending
2. **Adjust research depth**: Start with `quick`, upgrade to `standard` or `deep` as needed
3. **Refine team filter**: Add domains/emails to skip internal meetings
4. **Review briefs**: Check data/sent/ logs to see what's being sent
5. **Iterate on prompts**: Edit generate_brief.js to customize AI prompt

## Troubleshooting

**No briefs sent:**
- Check gcalcli authentication (`gcalcli agenda today tomorrow`)
- Verify calendar has events with external attendees
- Check logs in `logs/run.log`

**Briefs missing information:**
- Increase `research_depth` in config.json
- Check web_search and GitHub CLI are working
- Review research data in `data/research/`

**Duplicate briefs:**
- Check `data/sent/` for already-sent tracking
- Verify cron job isn't running multiple times

## Integration with OpenClaw

This skill uses:
- **gcalcli-calendar**: For fetching today's meetings
- **web_search**: For LinkedIn and company research
- **GitHub CLI (`gh`)**: For GitHub profile lookup
- **memory_search**: For past interactions (deep mode)
- **gmail skill**: For sending brief emails
- **sessions_spawn**: For AI brief generation
- **cron**: For daily scheduling

## Privacy & Security

- Research data is cached locally in `data/research/`
- No external APIs (uses web_search, GitHub CLI, memory_search)
- Briefs sent only to configured email
- Team member filtering prevents leaking internal info
- All data stored in skill directory (no cloud storage)