My latest prompt for my Claw. I thought this was pretty great so I thought I would share: Everyday at 4am, go read through the previous 24 hours of chat, and find all of the ways that we could improve our process. Figure out how we can improve those processes and write code that will improve it. These improvements might be a feature to add to siathinks, or a Claude Code command, or a whole new application. Lean toward the Claude Code command, and that could mean updating an existing command. Write the results down into codee-thinks directory in your codee-claude repo. The results should be actionable, something that if I say, build that, you and your sub-agents will be able to build. #openclaw

99 little bugs in the code, 99 little bugs. take one down, code it around, 137 little bugs in the code #life

Creating Claude commands to make life better. Currently creating a vibe-to-structure.md that takes a "vibe" written instruction and turns it into a structural construct that an LLM will use to produce better code #life

Just discovered that an article I wrote in 2003 was used in a presentation in 2004 in a completely different country. #life

My claw is researching the competition SEO. #winning

My Claw wrote some truly amazing copy today. 🧡

Will you let your Claw post on Moltbook? #life

When the LLM tells you that you were smart everytime... it makes it less impactful when you actually come up with a creative solution for it build. #life

Moltbot installed. #life

Is it better to add new functionality or perfect the features that exist? #life

Drove to Dad's house. #life

Not enough sleep #life

I broke user authentication. That's what I shipped. #bees

Writing a Claude Code agent to handle creating worktrees and branches and postgres databases oh-my #life

The key is being able to describe the end-state.

Streak saved! :tada:

Dang I lost track of time and there goes my streak lol

Im probably going to switch to SQLite for my Rails work now. AI and Worktrees and Postgres are not a great mix. Unless I can figure out how to create a Postgres DB based on a branch name. Anyone else have this figured out?

New LLM Workflow: Giving the same "idea" prompt to Gemini and Humanlayer (which uses Claude, but is Humanlayer's research prompt) to create two "research" documents. Will then hand both research documents and the idea prompt to Claude to create a PRD. Will then ask Claude to double check that we have all of the requirements gathered (this is critical, it has been found that the LLMs drop requirements frequently). Will take that final PRD and create Beads. Will then set Claude Code loose to write the code from the Beads. #life

The 37Signals guidance of "build the UI first" has always annoyed me, a) because it usually leads to the correct UI, b) as a backend dev I like making the models! First version of the new project is a mess of dashboardy looking nonsense.