Claude, tell me what needs my attention today
Each morning I open four tabs: Calendar, Gmail, Jira, and Obsidian. I read all, discard low priority emails or calendar events, and start building a mental picture of what I’ll do that day, and when exactly. For each piece of data I decide whether it’s time sensitive or not, whether I need to prepare beforehand, and whether I should tackle it later (or never). After warming up my brain, I kick off the actual work.
I decided an LLM should do that pre-work instead of my well-rested brain. To build such automation I’d practice Claude Code subagents and local MCP servers setup,
a good exercise for my new startup, RailsPilot.ai. So I started creating my /today Claude Code command.
/today fetches information scattered accross mails, calendar, task manager, and notes, and compiles them into a prioritized task list for the day.
It saves the output to a plain text file like daily_notes/2025-12-23.txt).
It’s funny to see Claude properly weighing what’s worthy of my attention from disparate sources according to natural language explanations, while it juggles different commands to decide what day is today. Discovering what’s cheap vs challening with LLMs is what I seek, and is also funny to me. December 22nd, 2025 is NOT Sunday, Claude, it’s Monday. Wise up!
It first works to define what day is today, then launches three parallel subagents, each responsible for fetching, processing and normalizing data from one source:
- The Jira subagent fetches assigned issues and checks for QA comments that need my attention: my highest priority these end-of-year days.
- The Calendar subagent retrieves today’s events and applies priority logic (marks certain family events as low priority reminders, whereas time-sensitive meetings are marked as high priority and with an action item to prepare 15 minutes beforehand).
- The Gmail subagent reads my inbox, detects newsletters to deprioritize and defines suggested action items for the rest.
The architecture is designed for efficiency: raw API responses are contained within each subagent’s context window and discarded after processing, so only compact JSON summaries flow back to the main agent. This keeps the context clean and allows each source to fail independently without breaking the entire command. It will also let me enable only relevant MCPs per subagent, another context window optimization I didn’t yet need.
Each subagent normalizes its data into a common task schema with fields like title, source, priority, due date, URL, and action_item. Low-priority items include a reason why. This normalization allows disparate data pieces to be compared and sorted. When all subagents return their results to the orchestrator agent, it merges everything into a single list sorted by priority and time.
The final output is plain text designed for quick scanning top to bottom, with key details, URL and action item after each title.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Daily Tasks - December 23, 2025
==============================
• [DEV-450] Enable PWA On Select Pages
- Status: IN QA
- Action: Respond to QA comments from David regarding errors on refresh
- URL: https://client.atlassian.net/browse/DEV-450
• Your Documents are now ready for Reservation ID: 123456
- From: Provider <^DONOTREPLY@provider.com>
- Action: download and print documents, complete forms
- URL: https://mail.google.com/mail/u/0/#inbox/123456789
• [DEV-430] Landing Page N+1s
- Status: IN QA
- Action: No QA feedback yet
- URL: https://client.atlassian.net/browse/DEV-430
• Kids visit Granny - 3:00 PM to 7:00 PM
- Note: Family reminder
- URL: https://www.google.com/calendar/event?eid=eideideideideideideid
---
Generated at 2025-12-23 09:30:00
And, instead of sifting through tabs, I can go from cup of coffee straight to my top priority for the day.