/time-audit — Weekly Time Audit

Run a personal time audit using macOS Screen Time data and Chrome history. Produces an HTML visualization showing where time actually went and how to improve.

Instructions

Step 0: Pre-flight check

Run these in parallel:

1. Verify Full Disk Access: sqlite3 ~/Library/Application\ Support/Knowledge/knowledgeC.db ".tables" 2>&1
2. Check system timezone: date
3. Find all Chrome History files: find ~/Library/Application\ Support/Google/Chrome/ -name "History" 2>/dev/null

If knowledgeC.db returns a permissions error, stop and tell the user:

"Terminal needs Full Disk Access to read Screen Time data. Go to System Settings → Privacy & Security → Full Disk Access → enable Terminal. Then re-run /time-audit."

Step 1: Get context (REQUIRED — do not skip)

Before pulling any data, ask the user these questions in a single message:

Quick context before I pull the data:

1. What's your role and what does a good week look like for you? (e.g. "GM of Courses — good week = 3+ hours of deep work daily, Slack under control")
2. What are you hoping to learn or fix? (e.g. "I think I'm spending too much time reactive", "I want to see if weekends are bleeding into work", "just curious")
3. Any apps or sites you already know are time drains you want to flag?

Wait for the user's answers before proceeding. These answers frame all findings.

Step 2: Pull knowledgeC data

Run in parallel:

Query A — Weekly app totals:

SELECT
  ZOBJECT.ZVALUESTRING as app,
  ROUND(SUM(ZOBJECT.ZENDDATE - ZOBJECT.ZSTARTDATE) / 3600.0, 2) as total_hours,
  COUNT(*) as sessions
FROM ZOBJECT
WHERE ZOBJECT.ZSTREAMNAME = '/app/usage'
  AND ZOBJECT.ZSTARTDATE > (strftime('%s', 'now') - 978307200 - 7*86400)
  AND ZOBJECT.ZVALUESTRING IS NOT NULL
GROUP BY app
ORDER BY total_hours DESC;

Query B — Daily totals with correct day labels:

SELECT
  date(ZOBJECT.ZSTARTDATE + 978307200, 'unixepoch', 'localtime') as day,
  strftime('%A', ZOBJECT.ZSTARTDATE + 978307200, 'unixepoch', 'localtime') as weekday,
  ROUND(SUM(ZOBJECT.ZENDDATE - ZOBJECT.ZSTARTDATE) / 3600.0, 2) as total_hours
FROM ZOBJECT
WHERE ZOBJECT.ZSTREAMNAME = '/app/usage'
  AND ZOBJECT.ZSTARTDATE > (strftime('%s', 'now') - 978307200 - 7*86400)
GROUP BY day
ORDER BY day DESC;

Note: knowledgeC timestamps use CoreData epoch (Jan 1, 2001). Add 978307200 to convert to Unix. Use localtime — never hard-code a UTC offset.

Step 3: Pull Chrome history

Find the active profiles: Copy each History file to /tmp, then compare MAX(last_visit_time) across all profiles. Only query profiles with data from the last 7 days.

for profile in "Default" "Profile 1" "Profile 2" "Profile 3" "Profile 4" "Profile 5" "Profile 6"; do
  db="/Users/$(whoami)/Library/Application Support/Google/Chrome/$profile/History"
  cp "$db" "/tmp/chrome_${profile// /_}.db" 2>/dev/null
  result=$(sqlite3 "/tmp/chrome_${profile// /_}.db" "SELECT datetime(MAX(last_visit_time)/1000000 - 11644473600, 'unixepoch', 'localtime') FROM urls;" 2>/dev/null)
  name=$(cat "/Users/$(whoami)/Library/Application Support/Google/Chrome/$profile/Preferences" 2>/dev/null | python3 -c "import json,sys; d=json.load(sys.stdin); print(d.get('account_info',[{}])[0].get('email','unknown'))" 2>/dev/null)
  echo "$profile ($name): $result"
done

For each active profile (last visit within 7 days), run the site categorization query. Adapt the CASE statement categories to match the user's role and tools — add/remove categories based on the context gathered in Step 1.

Chrome timestamps: last_visit_time / 1000000 - 11644473600 = Unix timestamp. Chrome has visit counts, not time — treat as activity proxy.

Standard category template:

SELECT
  CASE
    WHEN url LIKE '%mail.google.com%' THEN 'Gmail'
    WHEN url LIKE '%docs.google.com%' THEN 'Google Docs/Sheets'
    WHEN url LIKE '%drive.google.com%' THEN 'Google Drive'
    WHEN url LIKE '%calendar.google.com%' THEN 'Google Calendar'
    WHEN url LIKE '%meet.google.com%' THEN 'Google Meet'
    WHEN url LIKE '%notion.so%' OR url LIKE '%notion.com%' THEN 'Notion'
    WHEN url LIKE '%slack.com%' THEN 'Slack (web)'
    WHEN url LIKE '%linkedin.com%' THEN 'LinkedIn'
    WHEN url LIKE '%youtube.com%' THEN 'YouTube'
    WHEN url LIKE '%claude.ai%' OR url LIKE '%anthropic.com%' THEN 'Claude.ai'
    WHEN url LIKE '%github.com%' THEN 'GitHub'
    WHEN url LIKE '%zoom.us%' THEN 'Zoom'
    WHEN url LIKE '%localhost%' OR url LIKE '%127.0.0.1%' THEN 'Localhost (dev)'
    ELSE 'Other'
  END as category,
  COUNT(*) as visits,
  date(last_visit_time/1000000 - 11644473600, 'unixepoch', 'localtime') as day
FROM urls
WHERE last_visit_time/1000000 - 11644473600 > strftime('%s','now') - 7*86400
GROUP BY category, day
ORDER BY day DESC, visits DESC;

Also pull top domains in "Other" to identify unlabeled sinks.

Step 4: Generate HTML visualization

Save to a sensible path — suggest ~/Desktop/time-audit-{month}{year}.html or wherever the user typically works.

The HTML should include:

1. 4 stat cards — Total screen time, Deep work hours (terminal/focused tools), top comms app sessions, weekend hours
2. Daily bar chart — Color-coded: work days (dark green), personal days (red), mixed (amber), no data (gray), partial (light green). Use actual day names from data — never assign manually.
3. App totals — Horizontal bar chart, all apps from knowledgeC
4. Chrome breakdown — Work profile vs personal profile side by side, with visit bars
5. Where work time goes — Ranked estimated hours per activity
6. Findings — 4–6 cards, color-coded green/yellow/red, framed against the user's stated goals from Step 1

Design: clean, minimal, system font, white cards on light gray background. No dependencies — pure HTML/CSS.

Open the file automatically after writing: open <path>

Step 5: Deliver findings

Frame every finding against what the user said in Step 1. Lead with the most surprising or actionable insight, not a summary of what the data shows.

Structure:

1. The one thing that stands out most (1 sentence)
2. 3–4 specific findings, each tied to the user's stated goals
3. 2–3 concrete recommendations — specific and behavioral, not generic ("batch Slack to top of hour" not "use Slack less")

Notes

• Always derive day-of-week from the date — never assign manually. Use strftime('%A') in SQL or Python's date.strftime('%A').
• Chrome history may be locked if Chrome is open. Always cp to /tmp first.
• knowledgeC session count is often more revealing than total time — flag high session counts (200+/day on a single app) as fragmentation signals.
• If a profile has a personal Google account, label it clearly and call out that personal browsing will be visible — let the user decide how much detail they want.
• The HTML output is self-contained and shareable — mention this to the user.