Sync Meetings

Complete pipeline to sync Fathom meetings and organize them in workspace/meetings/.

Prerequisites

• FATHOM_API_KEY configured (see skill fathom)
• jq installed
• Script fathom.sh available at .claude/skills/fathom/fathom.sh

Full Workflow

When triggered, execute the steps below in order:

Step 1 — Fetch today's meetings

By default, fetch only today's meetings:

# Fetch today's meetings with summary and action items
{project-root}/.claude/skills/fathom/fathom.sh meetings --after "$(date +%Y-%m-%d)" --include-summary --include-actions

If the user specifies a different period (e.g., "sync this week", "sync yesterday"), adjust --after and add --before as needed:

• "sync de ontem": --after "$(date -v-1d +%Y-%m-%d)" --before "$(date +%Y-%m-%d)"
• "sync da semana": --after "$(date -v-7d +%Y-%m-%d)"
• "sync do mês": --after "$(date -v-1m +%Y-%m-%d)"

The API already returns default_summary.markdown_formatted and complete action_items — no extra calls needed.

Step 2 — Filter unprocessed (CRITICAL — anti-duplication)

Read the file of already processed IDs:

{project-root}/workspace/meetings/.state/fathom-processed-recording-ids.txt

Compare with the returned recording_id. Process only IDs that do not exist in this file.

IMPORTANT: This step is mandatory and cannot be skipped. If the ID already exists in the file, the meeting MUST NOT be reprocessed under any circumstances — no summary, no tasks, no notification.

If there are no new meetings, send via Telegram "🎙️ Sync Meetings — No new meetings." and stop immediately. Do not continue to the following steps.

Step 3 — Save raw JSON

For each new meeting, save the complete JSON to:

{project-root}/workspace/meetings/fathom/YYYY-MM-DD/YYYY-MM-DD__{recording_id}__{slug-do-titulo}.json

Where:

• YYYY-MM-DD = date from created_at
• slug-do-titulo = title in lowercase, spaces→hyphens, without special characters

Step 4 — Classify project

Determine the project based on the meeting title:

Step 5 — Save summary

Use the default_summary.markdown_formatted that already came in the API response (Step 1).

Read the template at .claude/templates/meeting-summary.md and fill with the meeting data.

Save to:

{project-root}/workspace/meetings/summaries/{project}/YYYY-MM-DD__{project}__meeting__{slug}__{recording_id}.summary.md

File format (based on the template):

---
date: YYYY-MM-DD
title: {original title}
project: {project}
type: meeting
status: summary
tags: [fathom, meeting]
recording_id: {recording_id}
recording_url: {url or share_url}
people: [{names from calendar_invitees}]
---

{content from default_summary.markdown_formatted}

## Action Items

{list of action_items formatted as checklist:}
- [ ] **{assignee.name}** — {description} ([{recording_timestamp}]({recording_playback_url}))

Step 6 — Todoist triage (action items to tasks)

For each processed meeting, extract the action_items and create tasks in Todoist.

BEFORE CREATING ANY TASK — mandatory anti-duplication check:

1. Check the local state file:

   Copy

   {project-root}/workspace/meetings/.state/fathom-todoist-sync.json
   

   If the recording_id already has synced tasks, DO NOT create new tasks. Skip to Step 7.

2. Search Todoist for existing tasks with the meeting title or recording_id in the comment:

   bashCopy

   todoist list --filter "search: {meeting title}"
   

   If you find tasks that clearly correspond to the same action items, DO NOT duplicate. Record the existing IDs in fathom-todoist-sync.json and skip.

Triage rules (only if passed the check above):

1. Translate to PT-BR — all action items must be translated to Brazilian Portuguese
2. Default project: Evolution — all tasks go to the Evolution project in Todoist, unless explicitly instructed otherwise
3. Actionable context — each task must have:
   • Clear and translated title (do not copy raw English from Fathom)
   • Comment with concrete context: origin (meeting + date), objective, next step, and recording link
4. Group by meeting — use sections/labels to identify which meeting it came from
5. Filter by assignee — create tasks only for action items assigned to the user (or without assignee). Items assigned to others are recorded only in the summary as reference

Todoist task format:

Title: {translated and clear action in PT-BR}
Project: Evolution
Priority: p3 (default) — raise to p2 if blocker or near deadline
Comment: 
  📋 Origin: {meeting title} ({date})
  🎯 Objective: {what this action resolves}
  ➡️ Next step: {concrete action}
  🔗 Reference: {recording_playback_url link}

Execute directly, without intermediate report. Do not list tasks before creating — create and confirm at the end.

Step 7 — Mark as processed (IMMEDIATELY after each meeting)

CRITICAL: This step must be executed immediately after processing EACH meeting individually, NOT at the end of all. This prevents a crash mid-processing from causing reprocessing.

Add the recording_id to the state file:

{project-root}/workspace/meetings/.state/fathom-processed-recording-ids.txt

One ID per line. Append, do not overwrite.

Also update fathom-todoist-sync.json with the created task IDs.

Order per meeting: Step 3 → 4 → 5 → 6 → 7 (write state) → next meeting.

Step 8 — Final report

When finished, present a short summary:

## Sync Fathom — Completed

**Period:** {oldest date} → {most recent date}
**New:** {N} meetings processed
**Already processed:** {M} skipped
**Tasks created:** {T} in Todoist (Evolution project)

### Synced meetings:
| Date | Title | Project | Tasks |
|------|--------|---------|---------|
| ... | ... | ... | {N tasks} |

Without listing tasks one by one — just counts. If the user wants details, they ask.

Step 9 — Notify via Telegram

Send the Step 8 summary via Telegram to the user using the /int-telegram skill:

• Chat ID: YOUR_CHAT_ID
• Use reply(chat_id="YOUR_CHAT_ID", text="...") via MCP
• Short format: emoji + title + meeting and task count

If there are no new meetings (stopped at Step 2), send: "🎙️ Sync Meetings — No new meetings."

Notes

• Do not reprocess meetings that already exist in fathom-processed-recording-ids.txt
• If the transcript is not available in the API (empty return), save the summary anyway and mark the transcript as status: pending
• Create directories automatically if they do not exist (fathom/YYYY-MM-DD/, raw/{project}/, summaries/{project}/)
• Maintain the existing naming convention — check examples in raw/ and summaries/ before saving
• Todoist triage: translate to PT-BR, Evolution project, actionable context, execute without intermediate report
• User's tasks only — other people's action items stay only in the summary
• Always use pt-BR in status messages