log-steam-playtime

name	log-steam-playtime
description	This skill should be used when the user asks to "track Steam playtime", "set up Steam playtime logging", "capture daily Steam hours", "parse Steam session logs", "set up a Steam Fabric pipeline", "create a Steam playtime notebook", "log game session history from Steam", "build a Direct Lake model on Steam data", or "ingest Steam stats into a lakehouse". Covers local snapshot scripts, Microsoft Fabric cloud pipeline (3 Delta tables + Direct Lake semantic model), and historical session recovery from Steam log files including multi-machine setups (PC + Steam Deck).
version	0.3.0

Steam Playtime Logging

Track Steam gaming activity across three layers, all landing in a single Microsoft Fabric Lakehouse:

Table	Granularity	Source
playtime_snapshots	Full library, one row per (game × snapshot)	Daily Steam Web API call
playtime_deltas	Only games where playtime_total increased since last snapshot	Computed from prior snapshot
sessions_log	Per-session start/end/duration, multi-machine	Parsed from gameprocess_log.txt

A Direct Lake steam_playtime.SemanticModel exposes all three for reporting.

Local config

Before doing anything else, load the local config so the actual Steam ID, machine label, vault names, and lakehouse name are in context:

@.local/steam-config.md

If that file is missing, copy .local/steam-config.md.example to .local/steam-config.md and have the user fill it in. The .local/ directory is gitignored.

For full details on credential storage (1Password / Azure Key Vault / OS keyring), see references/api-key.md. For directory layout and Delta table schemas, see references/workspace-structure.md. For log file locations, format, and parsing rules, see references/log-files.md. For Steam Deck specifics (SSH setup, mesh-router gotchas), see references/steam-deck.md.

Prerequisites

• Steam Web API key (free at https://steamcommunity.com/dev/apikey) stored per references/api-key.md
• Microsoft Fabric workspace with a Lakehouse. Direct Lake requires an F2 (or higher) capacity; the daily notebook itself runs on any Fabric SKU
• Python 3.10+ (uv add requests keyring)
• fab CLI logged in (fab auth login)
• 1Password CLI (op) or az CLI if using those backends for the API key

1 — Local snapshots

scripts/snapshot.py captures a point-in-time JSON + CSV of all owned games. It reads STEAM_ID and STEAM_API_KEY from the environment; falls back to the OS keyring (service=steam, username=api_key) if STEAM_API_KEY is unset. Do not edit tracked source — keep config in .local/steam-config.md.

Run a snapshot:

export STEAM_ID=<your-steam64-id>
# Pick one — see references/api-key.md for all options
export STEAM_API_KEY=$(op read "op://Private/Steam Web API/credential")  # 1Password
# or rely on keyring (no env var needed)

python snapshot.py
# Writes snapshots/YYYY-MM-DDTHH-MM-SSZ.{json,csv}

Compare two local snapshots with scripts/delta.py:

python delta.py           # last two snapshots
python delta.py --latest  # same
python delta.py old.json new.json

Always use playtime_total = playtime_forever + playtime_disconnected — playtime_forever alone misses offline / Steam Deck time and will not match the Steam UI.

2 — Fabric cloud pipeline

2a — Provision workspace + lakehouse

The lakehouse name comes from .local/steam-config.md (steam_lh is the default in this skill).

fab ls .capacities
fab mkdir "steam.Workspace" -P capacityName=<your-capacity>
fab mkdir "steam.Workspace/steam_lh.Lakehouse"
fab get "steam.Workspace/steam_lh.Lakehouse" -q "properties.sqlEndpointProperties"

This skill assumes the legacy (non-schema-enabled) Lakehouse — tables at Tables/<name>, exposed via SQL endpoint as dbo.<name>. See examples/lakehouse/README.md.

2b — Daily notebook (get_steam_playtime)

Scheduled daily. Two writes per run (see examples/notebook_daily.py):

1. Append a full snapshot row per game to playtime_snapshots
2. Compute the delta vs the previous snapshot; append rows where playtime_total increased to playtime_deltas

On the very first run every game appears in playtime_deltas because there's no prior snapshot.

2c — Manual baseline (get_steam_total_playtime)

Same as the daily notebook but only the snapshot half. Use for an extra checkpoint outside the schedule. See examples/notebook_baseline.py.

2d — Sessions reload (load_sessions_log)

See examples/notebook_load_sessions_log.py. Globs every *.json in Files/sessions/, unions the rows, overwrites sessions_log. Idempotent — re-running with new JSONs just refreshes the table.

2e — Import notebooks

Generate .Notebook directories with Python (never PowerShell here-strings — encoding issues on Windows). Write .ipynb via json.dumps(nb, ensure_ascii=True) + encoding="utf-8":

fab import "steam.Workspace/get_steam_total_playtime.Notebook" -i ./get_steam_total_playtime.Notebook -f
fab import "steam.Workspace/get_steam_playtime.Notebook" -i ./get_steam_playtime.Notebook -f
fab import "steam.Workspace/load_sessions_log.Notebook" -i ./load_sessions_log.Notebook -f

2f — Schedule the daily notebook (automatic deltas)

The schedule on get_steam_playtime is what makes deltas land automatically every day — no local cron required:

WS="steam.Workspace"
NB="get_steam_playtime"
START=$(date -u +"%Y-%m-%dT06:00:00Z")
END="2099-12-31T06:00:00Z"

fab job run-sch "$WS/$NB.Notebook" \
  --type Daily --interval 1 \
  --start "$START" --end "$END"

Both --start and --end are required; omitting either causes [NotRunnable].

Verify the schedule:

fab job run-sch "$WS/$NB.Notebook" --list

Check the most recent runs (status, duration, fail reason):

fab job run-list "steam.Workspace/get_steam_playtime.Notebook"

Alternative: local cron / systemd timer

If running snapshots locally instead of (or in addition to) the Fabric schedule, wrap scripts/snapshot.py in cron / systemd / Windows Task Scheduler. Then upload to Files/snapshots/ and trigger get_steam_playtime:

fab cp ./snapshots/<latest>.json "steam.Workspace/steam_lh.Lakehouse/Files/snapshots/" -f
fab job run "steam.Workspace/get_steam_playtime.Notebook"

3 — Direct Lake semantic model

Build a Direct Lake model over all three tables in one shot:

python scripts/build_direct_lake_model.py \
    --workspace "steam.Workspace" \
    --lakehouse "steam_lh.Lakehouse" \
    --model-name "steam_playtime" \
    --tables playtime_snapshots playtime_deltas sessions_log \
    --schema dbo \
    --out ./steam_playtime.SemanticModel

fab import "steam.Workspace/steam_playtime.SemanticModel" -i ./steam_playtime.SemanticModel -f

The script resolves the lakehouse SQL endpoint host + ID via fab get, pulls each table's schema via fab table schema, and emits model.tmdl, expressions.tmdl, database.tmdl, definition.pbism, .platform, and one tables/<name>.tmdl per table. See examples/semantic_model/ for the resulting layout.

4 — Historical session recovery

Per-session reconstruction from gameprocess_log.txt. Locations, format, rotation behaviour, and edge cases live in references/log-files.md. Steam Deck specifics (KDE Connect vs SSH, mesh-router fixes, password recovery) live in references/steam-deck.md.

python scripts/parse_sessions.py \
    --machine-id <MACHINE_ID> \
    --log-path "<platform-specific path to gameprocess_log.txt>" \
    --format json \
    --output sessions_historical_<MACHINE_ID>.json

Upload to the lakehouse and reload sessions_log:

fab cp ./sessions_historical_<MACHINE_ID>.json \
       "steam.Workspace/steam_lh.Lakehouse/Files/sessions/sessions_historical_<MACHINE_ID>.json" -f
fab job run "steam.Workspace/load_sessions_log.Notebook"

Each additional machine = drop another JSON in Files/sessions/ and re-run the notebook. The notebook globs *.json so the table is multi-machine ready.

Edge cases handled by the parser:

• Multiple PIDs per session (DLSS updater, child processes): only the first adding PID per AppID marks session start
• No start (log truncated): session_start_local left blank, row still written
• Session open at end of file (game running): session_end_local left blank
• Sessions longer than min(8h, playtime_total_hours) are capped via duration_minutes_capped / duration_hours_capped (raw values preserved)

Resources

References (references/)

• api-key.md — get the Steam Web API key; store it in 1Password, Azure Key Vault, or the OS keyring
• workspace-structure.md — directory tree, Delta table schemas, semantic model layout
• log-files.md — Steam log file locations per OS, format, rotation behaviour, parser usage
• steam-deck.md — getting gameprocess_log.txt off the Deck (KDE Connect vs SSH), mesh-router and password-recovery gotchas

Examples (examples/)

• notebook_baseline.py — get_steam_total_playtime (manual / one-off)
• notebook_daily.py — get_steam_playtime (scheduled daily; snapshots + deltas)
• notebook_load_sessions_log.py — load_sessions_log (multi-machine reload)
• lakehouse/ — .platform + provisioning README
• semantic_model/ — Direct Lake TMDL skeleton

Scripts (scripts/)

• snapshot.py — local snapshot (JSON + CSV)
• delta.py — compare two local snapshots
• parse_sessions.py — parse gameprocess_log.txt to session CSV/JSON; takes --machine-id and --log-path
• build_direct_lake_model.py — generate the TMDL package for a multi-table Direct Lake semantic model
• pull_deck_logs.sh / pull_deck_logs.ps1 — scp Steam logs off the Deck via SSH key auth (reads host from keyring steam/deck_host); see references/steam-deck.md for one-time key setup

Local config (.local/)

• steam-config.md.example — template; copy to .local/steam-config.md (gitignored)