Production Solo developer

Command Center

The AI digital twin running behind the scenes

BashAppleScriptClaude CLIEleventyCloudflare Workers

Here's the honest reason Command Center exists: I'm not disciplined enough to check my calendar, review my reminders, triage my notes, prep for meetings, and plan my content — every single day — without forgetting something. Nobody is. The gap between "knowing what to do" and "actually doing it consistently" is where most productivity systems fall apart.

So I automated the consistency.

The Task System

14 Claude AI tasks run on a daily cycle. Each one is a Markdown file with YAML frontmatter that tells SONET when to run it, what model to use, what to read, and where to write:

name: Morning Briefing
schedule: Weekdays 8:45 AM ET
model: sonnet
reads: Calendar, Notes Inbox, dashboard-data.json
writes: Apple Notes "Command Center Daily Briefing"

The prompt body — that's the part that took months to get right. Not "summarize my calendar." More like: read these specific calendars, cross-reference with the reminders list, check if any prep notes exist for today's meetings, flag conflicts, and format it as structured HTML with generous whitespace because wall-of-text notes are useless at 6:30 AM.

The briefing is there when I wake up. The triage runs at 2 PM whether I remember or not. The weekly review happens Friday even when Friday is chaos.

The Content Pipeline

I'm running a 100-day LinkedIn challenge. That means content needs to move from idea to draft to published piece to atomized social posts — consistently. Here's the flow:

  1. Idea Bank — Captured thoughts, observations, frameworks. The raw material.
  2. Drafting — A Claude task generates drafts from the idea bank, calibrated to my voice. Not "write like Nick" — specific patterns. Opening hooks that start mid-thought. Frameworks with numbered lists. Closers that land on a single sentence.
  3. Atomization — Long-form articles get broken into LinkedIn posts, Instagram quotes, newsletter segments. The atomization prompt alone is 2,000+ words of voice calibration.
  4. Manual review — I still read everything before it goes out. The AI handles volume. I handle judgment.

The Dashboard

A PWA at eng2arc.com/cc — protected by Cloudflare Zero Trust — with draggable widgets showing today's events, a scoreboard (posts published, drafts ready, open reminders), and a milestone countdown.

Data flows from the tasks into dashboard-data.json, gets pushed to git, and Cloudflare Pages serves it. The dashboard is a read surface. The intelligence happens in the tasks.

The Infrastructure

12 Bash scripts. 5 launchd agents. Apple Calendar exports every 15 minutes. Outlook screenshots get processed into calendar events. Apple Notes get read and written via AppleScript. iMessage notifications with deep links to the relevant note.

It's not elegant. It's set -euo pipefail and timestamped log rotation and lock files for idempotency. The kind of work that doesn't make for exciting demos but means the system actually runs at 1 AM when nobody's watching.

What I Learned

The real value isn't the AI. It's the reliability. Claude provides the intelligence — the summarization, the cross-referencing, the drafting. But the thing that changed my daily operating rhythm was knowing that the check happened. Not hoping I'd remember. Knowing.

What's the thing in your workflow that you know you should do every day but only actually do when you remember?

Nick Armstrong Sales Engineering Manager, Cisco Federal