Build an AI Secretary That Finds Open Action Items and Plans Your Day (Works With Slack + Gmail + Calendar)

Create a new folder for this workflow, then open Codex or Claude Code inside that folder. You are not manually building the Mono note first. You are asking the agent to create the workflow, the files, and the reusable skill.

The starting prompt is simple:

Make me a skill that looks at my Slack, Gmail, and calendar every morning, prioritizes tasks high to low, and puts them at the top of MonoNote.md under today's date.

Every task should include feedback checkboxes for priority and relevance, plus status checkboxes for done, bump, and cancelled.

The prompt should be copy-pasteable. It needs to include the task source, the destination file, and the feedback checkboxes so the agent builds the right format from the start.

Next, tell the agent to create the two Markdown files this system needs:

• MonoNote.md for the running task list
• task-rules.md for priority and ignore rules

MonoNote.md can start blank. The agent will put the first real dated section at the top when you run the skill.

Then have it add any priority rules you have per task into task-rules.md. You can dictate this in plain English.

Start with priority rules like:

High priority means direct requests, same-day deadlines, meetings that need prep, blocked teammates, or anything where silence creates risk.

Medium priority means useful follow-ups, active project work, or replies that matter but do not block today.

Low priority means optional admin, loose research, maintenance, or things I only need to review when there is time.

Ignore newsletters, receipts, automated notifications, FYI-only threads, vendor outreach without a clear ask, and anything already resolved.

Keep this file short. It should help the skill make better calls, not become a giant personal operating manual.

Be stricter than you think at first. A short list you trust is better than a giant AI-generated backlog you have to clean up every morning.

Run the skill manually before you schedule anything. The first run should create today's dated section at the top of the Mono note and group tasks by priority.

Every task should include:

• the task itself
• a source label or link, such as Gmail, Slack, or Calendar
• status checkboxes for done, bump, and cancelled
• feedback checkboxes for priority and relevance

The source link is the trust check. If the agent says you need to reply to an email or prep for a meeting, you should be able to click back to the message, thread, note, or event that created the task.

After the list appears, mark a few items. Use done for finished tasks, cancelled for tasks that should not come back, and bump for anything that should roll into the next business day.

For training, mark items that were too high, too low, or not relevant. This is the part that makes tomorrow's list better.

If your agent asks before installing the skill globally, approve it only after you have reviewed the files it created.

Once the manual run works, ask Codex or Claude Code to create the weekday automation.

The automation should not just run the list generator from scratch. It should look at the training you gave yesterday, update the task rules guidance file, then run the full morning loop:

• Review yesterday's task section.
• Read the prioritization feedback.
• Check which tasks were marked cancelled.
• Check which tasks were marked Not relevant.
• Update task-rules.md from those signals.
• Find anything marked bump and move it into today.
• Scan Slack, Gmail, and Calendar for new tasks.
• Generate today's prioritized list at the top of the Mono note.

Use a prompt like:

Create a weekday morning automation for this skill.

Before generating today's list, review yesterday's Mono note section. Learn from:
- prioritization feedback
- tasks marked cancelled
- tasks marked Not relevant

Use that information to update task-rules.md. Then carry forward any task marked bump.

Then run the to-do list generation task for today using Slack, Gmail, and Calendar. Put the new list at the top of the Mono note under today's date.

Do not send messages, create calendar events, archive email, delete anything, or modify outside systems.

In our setup, the agent created the note, the rules file, the skill source, and an active weekday automation.

Keep the first version read-only outside of your own Markdown files. The agent can recommend replies, meeting prep, and follow-ups. It should not send or schedule anything until the list is consistently useful.

The system only gets better if you use the checkboxes.

At the end of the day, every task should be in one of three states:

• done
• bump
• cancelled

Then use the training checkboxes when the agent guessed wrong. If a task was noise, mark it not relevant. If something was ranked too low, mark the right priority. If you leave the training row blank, the automation can assume the task was already in the correct priority section.

That last rule saves time. You should not have to mark the same obvious high-priority task as high priority every day.

This fixes the biggest problem with AI task lists. Without feedback, the agent keeps rediscovering the same noisy emails and vague threads. With feedback, it learns what to ignore, what to carry forward, and what actually matters in the morning.