announcement-digest

Announcement Digest

Adapt for your source — swap the @[SOURCE] label, names, and classification rules.

This is a sanitized version of a working school-email digest skill.

Full walkthrough: https://claudeblattman.com/workflows/school-digest/

v1.0 — Announcement email digest skill

Compile a digest of high-volume announcement emails from a single source (school, employer, nonprofit, professional association). Extract what matters, skip the noise, send a summary.

Prerequisites

• Gmail MCP — for searching and reading emails
• Gmail label + filter — all source emails auto-labeled (e.g., @School, @CompanyNews)
• Classification rules — a config file with your calibrated rules (see walkthrough)
• send-email.py or equivalent — for sending the compiled digest

Config & State Files

• Rules: [YOUR_CONFIG_PATH]/digest-rules.md
• State: [YOUR_STATE_PATH]/digest-state.json

State file tracks: last_digest_date, digest_count, reported_message_ids, reported_message_ids_ttl_days (default 30).

Execution Steps

Step 1: Read State & Check Frequency

1. Read digest-state.json.
2. If last_digest_date is not null and fewer than 2 days have passed:
   • Output: "Nothing new since [last_digest_date]. Next digest available [date]."
   • Exit.
3. Store digest_count (needed for first-time intro check at Step 5).

Step 2: Fetch Labeled Emails

1. Compute lookback date: last_digest_date if set, otherwise 14 days ago.
2. Search Gmail: label:[YOUR_LABEL] after:[lookback_date in YYYY/MM/DD]
   • Do NOT use is:unread — labeled emails may or may not be marked read. Query by date range.
   • Page through all results (follow next_page_token).
3. Filter out any message IDs already in reported_message_ids.
4. If no new emails remain, exit with message.

Step 3: Fetch Email Content

1. For each new message, fetch full content via get_gmail_message_content.
   • Fetch in parallel batches of 10.
2. Group messages by Gmail thread ID. Keep only the most recent non-you message per thread.
3. Accumulate total word count across all fetched message bodies.
4. Prune reported_message_ids: Remove IDs older than reported_message_ids_ttl_days.