Planting @bot Seeds 🌱

My always-on assistant The Martin Bot, a Hermes Agent, reads my calendar, keeps my todo list and project notes, and sends me news briefings. Its main feedstock, however, are my daily notes.

The daily note cleanup nudge

I use Craft daily notes to capture links and jot down thoughts. Throughout the day, I dump links and random thoughts into the daily note. My agent has access to my notes via an MCP server. At 6:30 PM every day, a cron job runs that reads my daily note and flags things I should clarify or expand while they’re fresh in my mind.

Here’s an example from last Saturday.

This nudge from the bot gave me an idea … why don’t you look it up, bot!? What if I could leave instructions for The Martin Bot to clean up or enrich my notes while it reviews them? Something like, “@bot Tell me whether Radiohead used Max on In Rainbows or Kid A”.

The @bot instruction and #bybot tag

I already had a skill for parsing my daily notes. I added a section for processing @bot instructions.

### Bot instructions
Blocks beginning with "@bot" of the form "@bot <instructions>"
contain instructions that the agent must follow before answering 
any user questions about the content.
1. Read and execute the instruction. The agent may use any
   other available skills and tools to carry out the instruction. 
   For example, using the `todo-cli` skill to add todo items,
   or any other skill that matches what the instruction
   is asking for.
2. Delete the block "@bot <instructions>" and replace it with
   the result of executing the instruction, prefixing the result 
   with "#bybot "

Asking the agent to prefix its results with the #bybot tag is important. It doesn’t edit my notes outside of @bot instructions. My notes are my notes. I could ask the agent to just “make all of my notes better,” but I don’t want to become The Martin Bot. The agent’s additions are clearly separated and tagged.

Here are some examples of The Martin Bot‘s work from the past few days. Note that when you paste a URL on its own line, Craft turns it into a “rich link,” which is what you see in the screenshots below. Links shared to Craft from another iOS app make up most of my daily notes.

On the left is the raw note. The right is after the agent’s pass.

It can also follow instructions that require other skills or tools.

Ask now, enrich later

The idea and implementation is simple. But it feels like magic every evening when I get my daily note review and the @bot seeds have grown into little sprouts of knowledge.

Why don’t I just prompt a model directly and paste in the answer while I’m taking notes? It’s partly the tooling. While Craft does have an assistant in the app, it doesn’t have access to my agent’s skills. And it’s mostly that I’m in a note-taking flow and don’t feel like waiting for an answer and being tempted to edit it. It fits better into my end-of-day note processing routine, so I delegate it to The Martin Bot to handle in the future.

The bot contributions make my notes richer and become a permanent and searchable part of my knowledge base. Even the agent itself benefits when it sends me the daily Wikipedia article 30 minutes later at 7 PM. It selects an article that matches a theme from today’s notes.

Sometimes it decides to write a poem about the article.

And in case you were wondering too, Radiohead did use Max/MSP on In Rainbows, but not on Kid A.