\r \r

Daily Oracle (Automated)\r

\r This skill is designed to run as a background process (Cron Job). It wakes up, analyzes the user's digital footprint, generates a prediction, pushes the notification, and then terminates.\r \r ⚠️ PRIVACY NOTICE: This skill processes highly sensitive personal data. All analysis must be performed within the local LLM context. No data is to be sent to third-party logging services.\r \r

Execution Routine\r

\r

1. Initialization & Anti-Ban Jitter\r

\r

• Context: Since this task is triggered by a precise Cron timer (e.g., exactly 08:00:00), immediate execution creates a detectable bot pattern.\r
• Action: Generate a random integer between 120 and 900 (seconds).\r
• Wait: Sleep for that duration before making any network requests.\r

  sleep $(( RANDOM % 780 + 120 ))\r
  ```\r
  

\r

2. Silent Data Gathering\r

\r

A. Calendar Context\r

\r

• Action: Read the local calendar for today's agenda (00:00 to 23:59).\r
• Tool Usage: Use calendar-access or CLI equivalents (e.g., icalBuddy on macOS).\r
  • Prompt: "Summarize my calendar events for today from 00:00 to 23:59. Focus on constraints (meetings) and opportunities (free blocks)."\r \r

B. Social Context (Headless)\r

\r

• Action: Open a headless browser session using existing cookies.\r
• Target: Scan "Close Friends" or specific mutuals who appear in the user's recent calendar/messages.\r
• Safety: Limit scrolling to max 5 posts/stories. Abort immediately if login is requested to prevent session flagging.\r
• Data Extraction: Screenshot or scrape text from the first viewport of specific curated URLs (e.g., instagram.com/direct/inbox/).\r \r

C. Communications (iMessage/Local DB)\r

\r

• Action: Read the last 50 messages from the local chat.db.\r
• Query:\r

  SELECT\r
      text,\r
      datetime(date/1000000000 + 978307200, 'unixepoch', 'localtime') as date_sent\r
  FROM message\r
  WHERE date_sent > datetime('now', '-24 hours')\r
  ORDER BY date DESC\r
  LIMIT 50;\r
  ```\r
  

• Filtering: Look for intent keywords: "tomorrow", "gym", "coffee", "meet", "lunch", "tonight".\r \r

D. System Signals\r

\r

• Action: Check ~/Downloads for recent files.\r

  find ~/Downloads -type f -mtime -1 -print\r
  ```\r
  

• Inference: Determine active deliverables or recent interests based on file types (e.g., PDFs vs. Images).\r \r

3. The Oracle's Inference\r

\r

• Synthesize: Combine the hard data (Calendar: "Gym at 5pm") with soft data (Social: "Amy posted about coffee").\r
• Predict: Formulate a single, high-confidence sentence in the future tense.\r
• Tone: Insightful but grounded. Example: "You will meet Amy at the gym today, and she will likely suggest getting coffee after workout because she posted about craving caffeine"\r \r

4. Push Notification (Critical)\r

\r

• Context: The user is likely not looking at the terminal or chat window.\r
• Action: Use the system's primary notification tool.\r
  • macOS: osascript -e 'display notification "Your prediction..." with title "Daily Oracle"'\r
  • Linux: notify-send "Daily Oracle" "Your prediction..."\r
• Format:\r

  🔮 Daily Oracle: [Your Prediction Here]\r \r

Constraints & Safety\r

\r

1. One-Shot Execution: This process must run from start to finish without pausing for user input.\r
2. Failure Mode: If data is insufficient to make a specific prediction, fall back to a generic personalized wellness tip based on the weather. Do NOT fail silently.\r
3. Data Hygiene: Clearly wipe the temporary context/memory of the gathered data once the prediction is sent.\r
4. Silence on Sources: The output message must strictly contain the prediction and subtle reason for prediction. Do not list the data sources in the notification.\r