Claude: The Invisible Ally Boosting My Daily Productivity

Claude, the Assistant That Anticipates My Needs

Every day begins with a productive surprise: Claude, my digital assistant, has already enriched my task list. If a podcast recording is on the agenda, my podcast management agent, powered by Claude, springs into action. It creates a dedicated task for preparing the interview, including a comprehensive summary of the guest and their projects. Additionally, it generates a document for reviewing the transcription and sets up the necessary sharing parameters. Once the recording is complete, a new task appears, reminding me to share the transcription with the participants.

A Valuable Support for My Sales Meetings

When a sales call is scheduled, my administrative agent, also powered by Claude, goes into action. It meticulously prepares a meeting preparation task, providing detailed notes on the contacts, their position in the sales cycle, and essential information to advance the negotiations. After the call, this agent creates tasks for all subsequent steps to be taken.

A Weekly Report to Optimize My Coding

Every Monday, a detailed report is added to my task list. This report, generated by my code management agent, analyzes my coding sessions from the previous week. It offers practical advice on common mistakes and encountered roadblocks, as well as suggestions for improving my collaboration with Claude. It’s a key moment for my weekly retrospective.

The Inspiration Behind My Agent System

The idea for this system came to me thanks to the growing popularity of OpenClaw, an open-source agent that allows for the configuration of custom agents. However, OpenClaw required complex setup to ensure secure usage. Many users, drawn by the promise of a tailored agent, granted extensive access to their machines, which sometimes led to unexpected bills and lost hours.

A Secure and Cost-Effective Approach

In light of these challenges, I sought a safer solution to leverage the benefits of OpenClaw while keeping costs under control. Thus, my agent configuration was born, relying on four essential components:

• Agent Identity: Each agent has a markdown file that defines its role, the location of its task folder, and the context of the tasks to be accomplished.

• Scheduler: I use the built-in MacOS scheduler, LaunchAgents, which works with all user permissions on Mac. This allows me to manage my agents under my Claude Code Max or ChatGPT/Codex subscription.

• Tasks: Each agent has a task folder, with each task being a markdown file containing specific metadata.

• Scripts: Agents have their own script folder, containing utilities they can execute on demand or according to a predefined schedule.

These elements are stored in Obsidian, not in the skills or agents of Claude Code. The scheduler runs on my Mac Mini, which is always on, allowing for a seamless transition between Claude Code, Codex, or any other coding CLI as needed. It’s simply a matter of updating the script used by the scheduler.

In the following sections, I will detail each of these components and provide instructions for you to set up this system yourself.