Learn how to configure Claude Code as a proactive executive assistant that manages tasks, drafts communications, monitors your projects, and keeps you organized without constant prompting.
Most developers treat Claude Code as a coding tool. Type a prompt, get code, review and move on. That is leaving enormous value on the table.
Claude Code is a general-purpose agentic system. It can read files, browse the web, run scripts, access APIs, and take action on your behalf. Configured correctly, it operates as an executive assistant -- one that handles scheduling reminders, drafts communication, monitors external services, and surfaces the information you need before you ask for it.
This lesson covers how to build that assistant layer on top of your existing Claude Code setup.
An executive assistant does not wait to be asked for every small thing. They know your priorities, anticipate what you need, and handle the administrative overhead so you can focus on high-leverage work.
Translating this to Claude Code means:
All of this is achievable with the right combination of CLAUDE.md configuration, skills, and the /loop feature covered in Lesson 5.
The CLAUDE.md file is read at the start of every session. For executive assistant use, treat it as an onboarding packet -- the document your assistant reads before starting work each day.
Include:
- Your current priorities and active projects
- Communication preferences (who gets responses, what tone to use)
- Standing tasks (daily standup format, weekly report structure)
- External services and accounts the assistant should know about
- Decision-making frameworks (when to act vs. when to ask)
Keep it under 150 lines. A bloated CLAUDE.md consumes context that your assistant needs for the actual work.
Any task you do more than once should become a skill. Skills encode your preferences and working style so Claude Code does not have to rediscover them every session.
Useful skills for an executive assistant setup:
draft-standup -- reads recent git commits and Notion tasks, drafts a standup messageinbox-triage -- reads Slack messages or emails, categorizes by urgency, drafts repliesweekly-summary -- pulls data from your tracking systems, writes a weekly summaryproject-status -- checks a specific project or API endpoint and reports backBuild each skill iteratively. Run it, give feedback, let Claude Code update the skill document. After several iterations, the output becomes consistent enough to trust without heavy review.
The executive assistant becomes dramatically more capable when connected to your real data sources.
High-value MCP connections for this use case:
- Notion -- task lists, project boards, daily notes
- Slack -- message drafting and channel monitoring
- Google Workspace -- Gmail, Drive, Calendar (via the GWS CLI covered in Lesson 10)
- ClickUp or Linear -- project management and ticket tracking
With these connected, a single prompt like "check my open tasks and draft my standup" becomes executable rather than hypothetical.
The /loop feature (covered in Lesson 5) is what transforms Claude Code from reactive to proactive.
Practical executive assistant loops:
- Every 15 minutes: check ClickUp for new task assignments
- Every hour: review unread Slack mentions and flag anything requiring a response
- At a specific time: surface a reminder with context ("it is 3pm -- you have a call with [client] in an hour")
Set these up with natural language:
- "Every 30 minutes, check my Notion inbox and tell me if anything new came in"
- "At 4pm, remind me to send the weekly summary to the team"
Loops run for up to 3 days. The terminal or app must stay open. For true overnight monitoring, use scheduled tasks (available in the desktop app).
This is the most important configuration decision. An assistant that asks for approval on everything is just a slow chatbot. An assistant that acts on everything without checking is a liability.
Define the boundary explicitly in CLAUDE.md:
Auto-execute (no approval needed):
- Draft documents and save to a drafts folder
- Read and categorize incoming messages
- Run status checks and generate reports
- Update internal tracking files
Require approval before acting:
- Send any message or email
- Make any API call that creates or modifies external records
- Push code to a remote repository
- Take any action that is hard to reverse
This boundary makes your assistant trustworthy. Clear rules are better than vague instructions.
Here is what an executive assistant CLAUDE.md block might look like:
## Daily Workflow
- Morning: check open Notion tasks, draft standup, surface anything blocked
- During day: monitor Slack #general for mentions, draft replies in a holding folder
- End of day: summarize what was completed, what moved to tomorrow
## Communication Rules
- Drafts go to .tmp/drafts/ -- never send without my review
- Standup format: Done / In Progress / Blocked -- 3 bullets max per section
- Tone: direct and professional in work comms, casual in DMs
## Auto-execute Without Asking
- Reading and summarizing
- Drafting to .tmp/
- Status checks
## Always Ask First
- Sending anything
- Posting to any channel
- Deleting files
When the executive assistant layer is properly configured, a typical morning looks like:
The same work that used to take 20-30 minutes of context-switching across five different tools happens in one place.
Next lesson: Automate Social Media with Claude Code
HivePowered AI — AI Like a Pro Training