My current work involves a lot of context switching and incorparates a bunch of incoming information everyday: two Slack workspaces, email, calendar, internal P2s (blogs), and meetings. The most fragile piece in this chain is me. It’s a lot to keep up and and if I’m relying on me retyping it into a tracker or writing it into a task list, context is lost and relies on just me and my brain.
My brain is pretty useful, but if you add in my non-work obligations, there is a lot of space for human error.
I’d been using Claude in chat and Cowork to think through problems. It’s useful, but everything goes in one direction with me asking and Claude answering. What I needed was something that did the looking for me and helped me prioritize.
So I opened Claude Code for the first time.
And what ensued was a personal operations system that lives on my laptop built through conversations and no code I wrote myself.
| Piece | What it does |
| Obsidian vault | Single source of truth: TASKS.md, a memory file Claude maintains, rolling files for ideas / strategy / brainstorming captured from meetings. |
| Morning scan | Runs at 8:30 AM. Pulls Quill meeting transcripts, Slack mentions and self-DMs, Gmail for-Claude items, a curated set of internal P2 posts. Extracts tasks, projects, ideas, strategy. |
| Daily digest | Calendar, FYI items, stats, status. All written each morning, rendered at the top of the dashboard. |
| Web app | Localhost dashboard, two columns: actionable on the left, awareness on the right. |
| Watcher | Confirms the scan ran. If it didn’t, triggers a recovery agent so I always have a fresh queue by 10 AM. |
| Weekly drafts | Monday “what I’m working on” and Friday “what I shipped” posts for our cross-team thread. |
| Delegation | Helps me track what I have delegated to others, from the dashboard. |
Every task has buttons for pin, mark urgent, mark done, move to another section, delegate. Each click rewrites the source markdown directly.
I started with just the vaguest idea, I was looking to Claude Code to help me diagnose and plan. From there, each addition came from a frustration or a gap I noticed:
- Markdown task list, sectioned by strategic priority.
- Morning scan ingesting four sources, writing a queue.
- Auto-archive of ticked items into a dated Done section.
- Web app, so I could see everything without opening Obsidian.
- Watcher to catch silent failures of the scan.
- Per-section quick-add for inline capture.
- Section moves and delegation from the dashboard.
- Mobile access via Tailscale.
The first iteration was just an .md file in Claude. I had to tell Claude when I checked something off. Then we moved to Obsidian because that gave me a surface I could actually interact with and actually check things off my list. Then we moved to a browser page that does all the thinking so any action I take on the page is updated in the tool.
The most recent addition was that I realized I could be looking at this on my phone, which is sometimes less distracting since I wouldn’t have other things going on in the background. I had no idea how to accomplish this so I asked Claude and it gave me a suggestion to set up Tailscale — a free private mesh network that lets my own devices reach each other. Now the dashboard loads on my phone from a saved icon on my home screen. Amazing!
I didn’t write code. I described what I wanted and Claude built it. I said what felt wrong and Claude adjusted. When something broke, I described the symptom and Claude diagnosed. Amazing!
Leave a Reply