Thoughtful - Skill Directory

This directory contains scripts and documentation for the thoughtful skill.

Data is stored separately in ${WORKDIR}/thoughtful-data/ (defaults to ~/clawd/thoughtful-data/).

Structure

Skill directory: ${WORKDIR}/skills/thoughtful/

skills/thoughtful/
├── SKILL.md
├── README.md
└── scripts/
    ├── generate-summary.sh
    └── process-and-summarize.js

Data directory: ${WORKDIR}/thoughtful-data/

thoughtful-data/
├── config.json          # Your preferences and settings
├── state.json           # Processing state (last run, counts)
├── tasks.json           # Pending tasks, commitments, waiting-on items
├── people.json          # Relationship tracking per contact
├── summaries/           # Historical summaries (one per day)
└── context/             # Temporary processing data
    ├── recent-messages.json
    ├── recent-chats.json
    └── last-prompt.txt

Quick Start

1. Configure Whitelisted Groups

Edit config.json and add groups you want to track:

{
  "whitelistGroups": [
    {
      "jid": "120363421500795949@g.us",
      "name": "House party",
      "priority": "medium"
    }
  ]
}

To find group JIDs:

wacli-readonly groups list --json | jq '.[] | {jid: .JID, name: .Name}'

2. Generate Summary Manually

cd ~/clawd/skills/thoughtful/scripts
./generate-summary.sh

This will:

1. Fetch messages from wacli-readonly
2. Process and filter based on config
3. Generate prompt for LLM
4. Save to context/last-prompt.txt

3. Review the Prompt

cat ~/clawd/thoughtful-data/context/last-prompt.txt

This is what will be sent to the LLM to generate your summary.

4. Set Up Daily Cron

Create a cron job in OpenClaw:

{
  "name": "thoughtful-daily",
  "schedule": {
    "kind": "cron",
    "expr": "0 9 * * *",
    "tz": "Asia/Calcutta"
  },
  "sessionTarget": "isolated",
  "payload": {
    "kind": "agentTurn",
    "message": "Generate WhatsApp communication summary using the thoughtful skill",
    "deliver": true,
    "channel": "telegram"
  }
}

Configuration Options

Chat Filtering

{
  "includeAllDMs": true,          // Include all direct messages
  "whitelistGroups": [...],       // Only these groups
  "blacklistGroups": [...],       // Never these groups (overrides whitelist)
  "priorityContacts": [...]       // Always highlight these people
}

Tracking Features

{
  "trackSentiment": true,         // Detect tone shifts
  "trackResponseTime": true,      // Your reply patterns
  "trackRecurringTopics": true,   // Conversation patterns
  "trackCommitments": true,       // Promises you made
  "trackImportantDates": true,    // Events, deadlines
  "trackDecisions": true          // Choices you made
}

Summary Options

{
  "defaultTimeRange": "24h",
  "includeWaitingOn": true,
  "includePendingTasks": true,
  "includeRelationshipInsights": true,
  "communicationCoachMode": true,
  "tone": "warm, direct, emotionally intelligent"
}

Data Files Explained

tasks.json

Tracks action items across conversations:

• tasks: Things you need to do
• waitingOn: Things you're waiting for others to do
• scheduled: Meetings/events mentioned
• commitments: Promises you made
• decisions: Choices you made

people.json

Per-contact relationship tracking:

• Response time patterns
• Sentiment trends
• Conversation frequency
• Important notes

state.json

Processing metadata:

• Last processed timestamp
• Total messages processed
• Chats tracked
• First run flag

Manual Testing

Test Summary Generation

# Generate and view prompt
~/clawd/skills/thoughtful/scripts/generate-summary.sh
cat ~/clawd/thoughtful-data/context/last-prompt.txt

# Test with different time ranges
~/clawd/skills/thoughtful/scripts/generate-summary.sh "3 days"

View Recent Messages

wacli-readonly messages list --limit 50

View Chats

wacli-readonly chats list --limit 20

Search for Specific Keywords

wacli-readonly messages search "meeting" --limit 10

Integration with OpenClaw

The skill is designed to work with OpenClaw's LLM capabilities. The processing script:

1. Fetches and filters messages
2. Builds structured data
3. Generates communication coach prompt
4. (Future) Calls OpenClaw LLM API
5. Delivers summary via Telegram

For now, the prompt is saved to context/last-prompt.txt for manual testing.

Privacy & Security

• All data stored locally (no cloud sync)
• wacli-readonly is read-only (can't send messages)
• Scripts run in sandbox for isolation
• No external API calls except OpenClaw LLM

Troubleshooting

No messages found

• Check wacli-readonly is authenticated: wacli-readonly auth status
• Verify sync is working: wacli-readonly chats list

Groups not showing

• Add JID to whitelistGroups in config.json
• Get JIDs: wacli-readonly groups list --json

Script fails

• Check Node.js is installed: node --version
• Make scripts executable: chmod +x ~/clawd/skills/thoughtful/scripts/*.sh
• Run in sandbox: use proper exec call with host="sandbox"

Next Steps

1. Test manual summary generation
2. Review and adjust config
3. Set up daily cron
4. Integrate LLM API call in process-and-summarize.js
5. Add interactive task completion (Telegram buttons)

Philosophy

This isn't just a notification system - it's a communication partner that helps you:

• Remember what matters to people
• Show up consistently in relationships
• Communicate with intention, not just reaction
• Catch small things before they become big things

Your relationships deserve better than "sorry, forgot to reply."