In my last post, Revisiting the Memex, I gave the rundown of how I built my first agent.

I mentioned that I had Claude generate an agent-rules.md file that any new agent would have to read before doing anything.

The v1 of that doc was minimal; my primary objective was simply to get the first subroutine working. But after building a second agent this weekend, one that syncs my Substack posts to Obsidian, I knew it was time to beef up the agentic constitution. If the goal over the next few weeks was to build a set of subroutines, I needed a ruleset that documented my priorities, style, workflow, goals, etc.

So, I worked with Claude this morning and had a stellar ruleset written up in under 20 minutes. Again, as I mentioned in my Context Compounds post, the only reason it was so simple was because it had all the necessary information from my end from previous atomic ideas, commonplace notes, substack posts, and Readwise highlights. The only thing I had to do was clarify the following questions:

And now, the agent-rules.md file lives in my subroutine folder and each subroutine’s first instruction is to read the doc:

It’s worth pointing out that I’m definitely in a honeymoon phase with all this Claude Desktop-Obsidian-Agents-Subroutine workflow I’m figuring out right now. To keep from getting too ahead of myself or over-engineering the system, I’m treating this as a continuous experiment.

The mindset I’m embracing for this sprint is “hey I’m just identifying repetitive tasks in my workflow and trying to simplify them with one pager instruction sheets”. I don’t want to frame this like some crazy agent overlord is taking over my life. I’m just keeping it simple stupid.

I’m treating the agent-rules file as a live constitution. This isn’t a document to revisit once a quarter; instead, I’m feeding it relevant updates or context continually. To keep the loop closed, each subroutine’s last step is to provide feedback or high-level observations that help the entire system evolve.

With that being said, though it’s somewhat personal, I’m sharing the exact copy of my agent-rules.md file because it will help you understand the nuances I’ve included that make it work.

Overall, my take here is that even if you’re not planning to use agents immediately, it’s worth creating this ruleset doc because it forces you to start thinking about what it would even mean to use agents in your workflow and knowledge management systems at all.

Note: This is my first time trying a paywall. I don’t care too much about the money I get from this specific post. This is more so an experiment for me to see if there’s even demand from readers to implement parts of my workflow and/or if this is the right form factor for it. If anyone has feedback, you know you can always e-mail me!

If you end up using it, my suggestion would be to copy paste the prompt & file into GPT / Claude and use the prompt below. It includes:

1. My original agent-rules.md file as is.

2. 13 pre-generated questions that it will ask you to better understand your workflows, goals, etc. Use voice mode here and just talk with your LLM freely as if you were having coffee with a friend.

3. Instructions for your AI on how to create a similar file for you and make any agents you deploy useful and personal.

The Prompt

(Copy paste everything from here to the end of the page)

Here is a sample agent-rules.md file. The constitution that every AI agent reads before doing anything in a person’s workflow. I want you to help me create my own version of this document.

Your job:

1. Read through the entire file below carefully

2. Identify every section that is specific to the person and would need to be rewritten for me

3. Ask me a focused set of clarification questions so you can rewrite those sections for my life, goals, and workflow

4. Once I answer, produce my personalized agent-rules.md file in the same structure and spirit as the original

Here’s the file: