sinzin91

I have a slack workspace with claws running on different models (opus, codex, Gemini). The use cases are too long to enumerate but this goes way beyond just EA for me.

I have this in my CLAUDE.md:

  ## Session Memory & Recall                                                                                                                                    
  - When asked to remember, recall, or look up something from past sessions, use the `/search-sessions` skill                                                   
  - Start with index search; if no results, suggest `--deep` for full message content search
  - Use `--project` filter when the context is clearly tied to a specific project

I have a separate project for that use case, good idea to integrate it here though!

disclosure: I have a horse in this race

Beyond a certain scale, you can't keep a mental model of the entire system in your head. What matters then is accessing accurate, up to date information the moment you need it (troubleshooting an unfamiliar subsystem, making a cross-cutting change). Table stakes are IaC, APM, structured logging.

Code-generated docs sound great in theory, but a huge category of knowledge never lives in code (RFCs, deployment processes, how to get prod access). Humans have to write and maintain those. That requires a culture where people believe their effort matters (and ideally gets rewarded). Without that, docs rot regardless of tooling.

Then there's discovery. The docs often exist; they're just buried. RAG tools actually help here imo. When people can find what you wrote, you're more motivated to keep it accurate. As we increasingly rely on agents to tell us about our system, we're going to have to ensure the docs they're reading are not woefully out of date or inaccurate.

You should check out Blender MCP, which allows you to connect Claude Desktop/Cursor/etc to Blender as a tool. Still early days from my experiments but shows where it could go https://github.com/ahujasid/blender-mcp