| name | activitywatch:activity |
| description | Report real device usage from ActivityWatch. Covers per-app time, window titles, and active vs idle spans. Use when asked "what apps did I use", "how long was I in X", "what did I work on today", "how much was I active vs idle", or to mine usage patterns for automation. |
| argument-hint | [query] [--recent <1d|7d|4w>] [-n <limit>] |
| allowed-tools | ["Bash","Read"] |
Activity
Report how the device was actually used, from ActivityWatch's local capture. This skill only reads. It never writes to the db.
ActivityWatch runs aw-server-rust in the background (a LaunchAgent supervises it via aw-qt). It records the focused app and window title (aw-watcher-window) and active vs idle spans (aw-watcher-afk) into a SQLite db. DuckDB attaches that db read-only, which is safe while aw-server holds its own write connection.
Database
~/Library/Application Support/activitywatch/aw-server-rust/sqlite.db
Override with AW_DB (the wrapper resolves it, since DuckDB cannot expand ~ in ATTACH). If the file is missing, ActivityWatch is not installed or has not captured yet.
Querying
Run a named query with scripts/aw-query.sh. Each query supports --recent <duration> (12h, 1d, 7d, 4w) to scope to a recent window, and -n <limit> to cap rows.
${CLAUDE_SKILL_DIR}/scripts/aw-query.sh top-apps
${CLAUDE_SKILL_DIR}/scripts/aw-query.sh top-apps --recent 1d
${CLAUDE_SKILL_DIR}/scripts/aw-query.sh window-titles --recent 7d -n 40
${CLAUDE_SKILL_DIR}/scripts/aw-query.sh afk --recent 1d
The aggregate queries (top-apps, window-titles, afk) return durations both formatted (duration, e.g. 1h 23m) and raw (seconds) so results are readable and still sortable/summable. app-timeline lists individual focus events and returns raw seconds only.
Queries
- top-apps — total focused time per app, most first. The headline "what did I use" answer.
- window-titles — time per window title within each app. What you were actually doing, not just which app was open. Needs Accessibility granted to ActivityWatch.
- app-timeline — focus events in reverse-chronological order (when you switched to what, for how long). Pair with
--recent 1d to reconstruct a day.
- afk — active (
not-afk) vs idle (afk) time. How much of the window was actually at the keyboard.
Schema
Two tables in the attached aw database:
buckets(id, name, type, client, hostname, created, data) — one row per watcher. type is currentwindow (window watcher) or afkstatus (afk watcher).events(id, bucketrow, starttime, endtime, data) — bucketrow joins to buckets.id. starttime/endtime are nanoseconds since the Unix epoch (divide by 1e9 for epoch seconds). data is JSON: {app, title} for currentwindow, {status} for afkstatus.
Duration of an event is (endtime - starttime) / 1e9 seconds. The named queries live in resources/queries/.
Discovery
For questions the named queries don't cover, attach the db read-only and explore. DESCRIBE and a sample surface any watcher's shape (browser, editor, and custom watchers all land in events.data as JSON):
duckdb -c "INSTALL sqlite; LOAD sqlite; INSTALL json; LOAD json;
ATTACH '$HOME/Library/Application Support/activitywatch/aw-server-rust/sqlite.db' AS aw (TYPE sqlite, READ_ONLY);
SELECT type, count(*) FROM aw.buckets GROUP BY 1;
DESCRIBE aw.events;
SELECT data FROM aw.events LIMIT 5;"
Write ad-hoc queries in the same shape as the files in resources/queries/: join events to buckets, filter by buckets.type, extract fields with json_extract_string(data, '$.field'), and measure time as (endtime - starttime) / 1e9.
