Extract action items from today's Granola/Grain meetings, create Todoist tasks, complete fulfilled tasks, and draft meeting-triggered follow-up emails. Use when running the daily action items cron, post-meeting cron, or when user asks to process meeting action items. NOT for general email drafting without meeting context.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Extract action items from today's Granola/Grain meetings, create Todoist tasks, complete fulfilled tasks, and draft meeting-triggered follow-up emails. Use when running the daily action items cron, post-meeting cron, or when user asks to process meeting action items. NOT for general email drafting without meeting context.
version
1.8.0
license
MIT
author
Martin Gontovnikas <martin@hypergrowthpartners.com>
Read ../config/user.json (resolves to ~/executive-assistant-skills/config/user.json).
Extract and use throughout:
name, full_name — to identify your action items in meeting notes (e.g. "Gonto (Martin)")
whatsapp — for result delivery
workspace — absolute path to OpenClaw workspace
Do not proceed until you have these values.
Debug Logging (MANDATORY)
Read ../config/DEBUG_LOGGING.md for the full convention. Use python3 {user.workspace}/scripts/skill_log.py action-items <level> "<message>" ['<details>'] at every key step. Log BEFORE and AFTER every external call (gog, mcporter, todoist-cli). On any error, log the full command and stderr before continuing.
Steps
0. Check today's calendar for meetings (BOTH accounts)
Before querying Granola, get today's actual meetings from BOTH calendars to know what to expect:
python3 {user.workspace}/scripts/skill_log.py action-items INFO "Starting action-items run"# Get today's date in YYYY-MM-DD and tomorrow's
TODAY=$(date -u -d "$(TZ=America/Argentina/Buenos_Aires date +%Y-%m-%d)" +%Y-%m-%d)
TOMORROW=$(date -u -d "$(TZ=America/Argentina/Buenos_Aires date -d '+1 day' +%Y-%m-%d)" +%Y-%m-%d)
# Check BOTH calendars
gog --account {user.primary_email} --no-input calendar list primary --from "${TODAY}T00:00:00-03:00" --to "${TOMORROW}T00:00:00-03:00" --json 2>&1
gog --account {user.work_email} --no-input calendar list primary --from "${TODAY}T00:00:00-03:00" --to "${TOMORROW}T00:00:00-03:00" --json 2>&1
Log the results: python3 {user.workspace}/scripts/skill_log.py action-items DEBUG "Calendar events found" '{"primary": N, "work": M, "total": N+M}'
Merge events from both calendars. Filter for meetings with attendees (skip solo/personal events). This gives you the ground truth of what meetings happened today — use it to cross-check Granola results and catch any meetings Granola missed.
CRITICAL date syntax: Use explicit ISO8601 dates with -03:00 offset. Do NOT use relative expressions like +1 day, today, or tomorrow in gog flags — they may not be supported. Always compute the actual date strings.
1. Get today's meetings from Granola
Timezone note: Granola stores meeting times in UTC. For ART (UTC-3), querying "today" means using today's date AND tomorrow's date in UTC. E.g., for March 3 ART, query custom_start: "2026-03-03" and custom_end: "2026-03-04" to capture all ART-day meetings.
Collect meeting IDs and titles. Log: python3 {user.workspace}/scripts/skill_log.py action-items INFO "Granola meetings found" '{"count": N, "titles": [...]}'
Cross-check with calendar: Compare Granola meetings against the calendar events from Step 0. If a calendar meeting with attendees has no Granola match (by time overlap within 15 min), log a warning — it may not have been recorded. Proceed with what Granola has, but note unmatched meetings in the output.
Skip if no meetings (from either Granola or calendar).
2. Query Granola for MY action items + email triggers
mcporter call granola query_granola_meetings --args '{"query": "What are all of {user.name} ({user.full_name}) personal action items, follow-ups, and commitments from these meetings? Only things HE needs to do, not what others committed to. For each item, include the specific person, company, project, or candidate name involved — never use generic references. Also identify: any promises made to do ANYTHING via email (intros, follow-ups, sending docs, sharing info, connecting people, etc.), and whether each meeting was a FIRST meeting with that person/company or a follow-up.", "document_ids": ["<id1>", "<id2>", ...]}'
Pass ALL meeting IDs. Preserve citation links.
3. Create Todoist tasks
Note: Load env in the same command — each shell call is a fresh session.
Read {user.workspace}/skills/todoist-api/SKILL.md for CLI usage. For each action item:
Meeting date and time (e.g. "Wed Mar 5, 15:00 ART")
Who requested / context
Granola citation link
Rules:
Only your actions: Granola summaries often list "next steps" without clear ownership. Be skeptical — if the action could belong to the other person (e.g. "digest their own content", "check availability", "get back to us"), do NOT create a task. When in doubt, create a FOLLOW-UP task ("Ping X about Y") rather than an ownership task.
Capture commitments from others: If the other person said they'd do something (e.g. "I'll get back in 2 days"), create a follow-up/ping task with the appropriate due date, not a task to do the thing yourself.
Specificity: Every task MUST include specific name of person/company/project
Due dates: If implied deadline, use --due with natural language
Labels: Tag: intro, follow-up, email, urgent as appropriate
Never create separate tasks for steps that are part of the same workflow. If the action is "prepare X then send X" — that's ONE task, not two. Examples:
❌ "Write draft" + "Send email" → ✅ "Draft and send email to X"
❌ "Review deck" + "Share deck" → ✅ "Review and share deck with X"
One task per intent. The user will naturally do the steps in order.
Todo dedup (MANDATORY)
Before creating a task, run a duplicate check against open Todoist tasks:
Normalize proposed title (lowercase, trim punctuation, collapse whitespace)
Search open tasks (source {user.workspace}/.env && todoist-cli list --filter "!completed") and compare normalized content
Treat near-identical intro tasks as duplicates (e.g., "Intro David to Marcos" vs "Intro David (n8n) to Marcos Nils")
If duplicate exists: do NOT create a new task; append meeting context to the existing task description when useful
In output, report dedup decisions under Skipped as duplicates:
4. Draft follow-up emails
For ANY email that needs drafting (intros, follow-ups, VC replies, sending docs, etc.):
Read and follow ~/executive-assistant-skills/email-drafting/SKILL.md — it is the single source of truth for all drafting rules, style, templates, humanization, and delivery
Identify draft triggers from meeting notes:
Promised intros or follow-ups
Promised docs/PDFs/decks
First call with a VC or new lead
Any promise to email someone
When in doubt about whether to draft → DRAFT IT
Exception: The proposal-only commitment rule below overrides this. If the commitment is to build a proposal first, do NOT draft — create a Todoist task only.
HGP Deck: {user.workspace}/assets/HGP_Deck_2025.pdf — attach via --attach {user.workspace}/assets/HGP_Deck_2025.pdf. Say "Hypergrowth Partners deck" in the email body (not "one-pager" or "our deck")
When drafting intros to known contacts: search sent emails for previous intros to them, use the same format, tone, and description
Intro-specific hard requirements (MANDATORY)
If action items include intros, follow this exactly:
Create one separate intro draft per intro pair (never merge multiple intros into one generic follow-up).
Subject must be explicit: Intro: <Person A> <> <Person B>.
Body must include:
one-line who each person is,
one-line context for why this intro is happening,
close with I'll let you two take it from here. and {user.signature}.
Never replace intro drafts with a generic recap email like "Great meeting today" when intros were promised.
If recipient emails are known, create Gmail drafts immediately; if any email is missing, still generate the full draft text and report MISSING_EMAIL: <name>.
In the WhatsApp result, include an Intro drafts created: section listing each intro pair and whether it was drafted in Gmail or blocked by missing email.
First VC / first dealflow call follow-up (MANDATORY)
When the meeting is a FIRST call with a VC or dealflow company:
Create a first-meeting follow-up draft (unless blocked by proposal-only rule below).
Use meeting-specific context in the body (what was discussed, explicit next steps, concrete offers), not generic pleasantries.
Include your positioning (how you work) and attach {user.workspace}/assets/HGP_Deck_2025.pdf when relevant.
Allowed to use "Great meeting today" only for this first-meeting follow-up class.
Proposal-only commitment rule (MANDATORY)
If you committed to "build a proposal" (or equivalent: proposal/deck/scope draft to prepare first):
Do NOT draft an outbound email yet.
Create a single Todoist task to build AND send the proposal — e.g. "Build and send advisory proposal to Gabriel (BairesDev)".
Do NOT create separate tasks for "build" and "send" — that's redundant. One task covers the full lifecycle.
5. Check if existing Todoist tasks were fulfilled in today's meetings
After processing action items, also check if any existing open Todoist tasks were addressed/completed during today's meetings:
List open tasks: source {user.workspace}/.env && todoist-cli list
For each meeting, check if the discussion covered or fulfilled any open task (e.g., "Share AI strategy with Colin" → discussed AI strategy directly with Colin in the call)
If a task was clearly fulfilled in the meeting, complete it: source {user.workspace}/.env && todoist-cli complete <task_id>
Report completed tasks in the output: "✅ Completed: [task] — fulfilled during [meeting name]"
This ensures Todoist stays clean and reflects what actually happened.
Error handling
If gog calendar fails: log the error with full command + stderr (python3 {user.workspace}/scripts/skill_log.py action-items ERROR "gog calendar failed" '{"command": "...", "stderr": "..."}'), continue with the other account and/or Granola-only.
If mcporter call granola fails with auth errors: report "⚠️ Granola auth expired, run mcporter auth granola --reset" and stop.
If todoist-cli fails: verify .env is sourced and token is valid. Report error.
If Grain MCP is unavailable: proceed with Granola-only extraction, note in output "⚠️ Grain unavailable, results are Granola-only."
If any step fails, continue with remaining steps when possible — don't abort the entire run for one failure.
FIRST STEP before any processing: Read {user.workspace}/state/processed-meetings-YYYY-MM-DD.json (if it exists — array of Granola meeting IDs)
Skip any meetings whose ID is already in that list — do NOT re-process them
Immediately after processing each meeting (before moving to the next), append its Granola meeting ID to the file. Do NOT wait until the end — write after each meeting to prevent races with other crons.
This file is the single source of truth for "was this meeting already processed today?"
Note: Titles are fragile for recurring meetings that share the same name. Always dedup by Granola meeting ID.
Task-level dedup
Before creating ANY task, check BOTH open AND recently completed tasks for near-duplicates:
Fetch open tasks: source {user.workspace}/.env && todoist-cli list
Fetch today's completed tasks: use Todoist Sync API completed/get_all with since=<today 00:00 UTC> or source {user.workspace}/.env && todoist-cli list --filter "completed today" if supported
Same person + same action intent = duplicate (e.g. "Text Morgane about Hank" and "Text Morgane (Braintrust) after Hank call")
If duplicate exists in EITHER open or completed → SKIP, do not create
Report skipped duplicates in output: "⏭️ Skipped (already exists): [task]"
Why check completed tasks: The user may have already completed a task created by a post-meeting cron earlier today. Recreating it is wrong — the work is done.
Why both layers matter
Post-meeting crons fire per-meeting. The daily end-of-day cron processes ALL meetings. Without meeting-level dedup, the same meeting gets processed twice. Without task-level dedup, even if the meeting is re-processed (e.g. file write failed), individual tasks won't be duplicated.
Dedup is NON-NEGOTIABLE
If a meeting appears in processed-meetings-YYYY-MM-DD.json, do NOT process it again under any circumstances — even if you think the post-meeting cron "might have missed something." The post-meeting cron already handled it. If it had issues, the user will ask for a re-run manually.
Output
Tasks created or drafts composed → list tasks with short summary
Tasks completed (fulfilled in meetings) → list with ✅
Always name the specific meetings processed (e.g. "from Fyxer call, Braintrust weekly, HGP Staff") — never say "from today's meetings"
No meetings or no action items → NO_REPLY
Cross-check with Grain (ACTIVE — transcript-level verification)
Grain MCP is available with full transcript access. After extracting action items from Granola:
Find the meeting in Grain: mcporter call grain.list_attended_meetings --args '{}'
Note: Grain's schema does not support start_date/end_date filters. Call with empty args and filter the results manually by start_datetime to match today's date range.
Fetch the transcript: mcporter call grain.fetch_meeting_transcript --args '{"meeting_id": "<grain_meeting_id>"}'
Cross-check each action item against the transcript:
Is the action item actually assigned to you, or to the other person?
Are there commitments with timelines that Granola missed (e.g. "I'll get back in 2 days")?
Are there action items Granola dropped entirely?
Prefer Grain's transcript over Granola's summary when they conflict
Also use grain.fetch_meeting_notes for Grain's own AI-generated notes as a second opinion
If a Grain meeting can't be found (e.g. recording wasn't on), fall back to Granola only + skepticism rules.
Matching Grain to Granola meetings
Match by time overlap (within 15 min of start time), not by title. Grain and Granola may name the same meeting differently. If no Grain match found within the time window, proceed without cross-check.
Overview
Extracts action items from daily meetings via Granola/Grain, creates Todoist tasks with deduplication, completes fulfilled tasks, and drafts follow-up emails triggered by meeting commitments.
Prerequisites
gog CLI configured with both Gmail/Calendar accounts
mcporter with Granola and Grain MCP connections authenticated
todoist-cli installed with valid API token in workspace .env
OpenClaw workspace with skill_log.py and audit_log.py scripts
WhatsApp delivery endpoint configured in user.json
Instructions
See the Steps section above (Steps 0 through 5) for the full execution workflow.
Examples
# Trigger via cron or manual invocation# The skill reads user.json, queries both calendars, pulls Granola meetings,# extracts action items, creates Todoist tasks, drafts emails, and notifies via WhatsApp.# Example output: "From Fyxer call: 2 tasks created, 1 intro draft (Gmail), 1 task completed"
