scheduled-summary

Scheduled Summary

Overview

Messaging platforms are excellent for one active conversation, but they hide the activity that happened elsewhere: other sessions, scheduled jobs, saved memory, and unresolved items from earlier work. This skill creates a periodic digest that surfaces that cross-session activity in a compact format suitable for chat messaging platforms or any notification channel that accepts Markdown or plain text.

The digest is designed for a cron scheduler. It reads optional local data sources (session store, cron output directory, memory JSON files, and log file), filters activity to a configurable time window, and emits Markdown, JSON, or plain text. When no data sources are configured, it emits a template digest that an agent can fill by querying its own platform-specific session history, scheduler status, and memory tools.

Quick start

From this skill directory:

python scripts/summarize.py generate --since 24h

With explicit data sources:

python scripts/summarize.py generate \
  --since 24h \
  --sessions-db /path/to/sessions.sqlite \
  --cron-dir /path/to/cron-output \
  --memory-dir /path/to/memory-json \
  --log-file /path/to/agent.log

Show current configuration and skipped sources:

python scripts/summarize.py config

Create a config template to fill in:

python scripts/summarize.py init --output .summary-config.json

What the digest includes

1. Session activity — sessions started or completed inside the time window, plus topic titles where available.
2. Cron job status — jobs that ran successfully, jobs that failed, and jobs marked overdue by their output files.
3. Memory changes — new saved memories and updated facts from JSON records.
4. Tool usage — most-used tools, total calls, and error counts from a text or JSONL log file.
5. Outstanding items — due decision reviews, incomplete tasks, and TODO-like items found in session transcripts.
6. Time window — every section is scoped to --since (24h, 7d, 2w, etc.) unless the source explicitly marks an item overdue.