Every morning I type one command and get a briefing that triages my email, checks my calendar, scans Slack, reviews yesterday’s meeting notes, and tells me which to-do items are going stale. It takes about 90 seconds and replaces what used to be 30-45 minutes of tab-switching.
I built this as a “skill” for Claude Code and CoWork. A skill is a reusable prompt that can orchestrate tools, manage state, and run autonomously. Less chatbot, more small program that happens to be powered by a language model.
The problem it solves
Every individual tool — email, calendar, Granola, Clay, Slack, task tracker — is fine on its own. The hard part is the synthesis. Knowing that an email from a customer relates to a priority I’m tracking, which connects to a meeting in two hours, which I should prep differently because of a Slack thread I missed last night. No single app does that. I was burning my best morning hour just assembling context.
How it works
On first run, it interviews you — your role, tools, priorities, important people. Then it asks you to rate a handful of sample tasks on a 1-5 scale and uses your ratings to calibrate a scoring model for how you think about urgency and importance. That’s the part I find most interesting. “Triage email with AI” sounds simple, but triage is deeply personal. Rather than hardcoding rules, the briefing learns a weighted model from your own judgments.
Each subsequent run kicks off parallel agents that scan your tools, cross-reference the results, and synthesize them into a single document. Items you’ve completed get filtered out. Items that keep appearing without being addressed get escalated.
Beyond the morning briefing
One skill doesn’t make a chief of staff. The daily briefing was the starting point, but the pattern turned out to be more general than I expected.
I built a board update skill that drafts the product section of our monthly investor update. It pulls from the same constellation of tools — project trackers, Slack, meeting notes, completed tasks — but the job is fundamentally different. Instead of “what should I pay attention to right now,” it’s “what story do we tell about this month.” It researches across sources in parallel, groups findings into a Progress / Priorities / Problems structure, and cross-validates every proper noun across at least two sources (AI-transcribed meeting notes get names wrong constantly). The output matches the institutional voice and format of updates I’ve written by hand for years.
There’s also a product brief skill. When my team is scoping a new feature, it pulls context from design files, customer research, support tickets, and existing specs, then runs a structured interview to fill gaps. It writes the brief, then reviews it from four perspectives — CEO, engineering, product, design — each with different concerns. The brief scales in depth based on how significant the initiative is, from a few paragraphs for a small fix to a full research-backed document for a strategic bet.
The common thread is the same: these are tasks where the gathering and synthesis dwarfs the actual thinking and writing. A good chief of staff doesn’t make decisions for you — they make sure you have the full picture when you sit down to make them.
What I learned
Building these taught me more about my own information diet than about AI. I was spending real time on emails where I wasn’t even the decision-maker. Carryover tracking forced honest conversations about whether recurring items were actual priorities or just felt like they should be. The “tool integration” part of AI agents is both less magical and more useful than people think. There’s no special intelligence in reading a calendar API. The value is in the opinionated synthesis layer on top — that’s where the language model earns its keep.
Why I open-sourced it
The daily briefing is on GitHub as a generic, self-configuring template that adapts to whatever tools you have connected. I don’t know if this specific implementation will age well — the tools, APIs, and capabilities will all change. But the underlying problem of synthesizing fragmented information into a coherent picture isn’t going anywhere. MIT licensed if you want a starting point for your own.
Leave a Reply