Agents have their own space!

I kicked the agent definitions out of the Mnemonic repo. They now have their very own git repository!

Why?

Because the agents were actively being used with my Cognee setup, and it was getting "clunky". I had to streamline it.

Back in the Mnemonic repo, agents where physically copied from the repo into $HOME/.claude/agents. Now, they're just symlinked. Just restart Claude Code and the changes are picked up. The repositoy is the source of truth!

I also refined the agents. I added some skills and extra tools to them:

skills:
  - arch-docs
  - mermaid-diagrams:mermaid-diagrams
  - writing-clearly-and-concisely:writing-clearly-and-concisely
tools:
  - "mcp__cognee__search"
  - "mcp__context7__resolve-library-id"
  - "mcp__context7__query-docs"

I also created my own code review skill and added it to my code reviewer agent. Yeah, there are some pretty well defined code review skills and prompts out there, but I have mine usin my agents that I've labored over the past 5+ months. They're starting to work the way I like.

The project still relies on Cognee for memory, but that is only the case until Mnemonic is ready...which will be soon! I anticipate that the MVP will be ready by the end of February.

Have a look at it, clone it and tweak it. Have fun with it!

Related: Claude Code Setup project page