Me when Claude Code goes down...

I’m tired of people saying that it’s impossible to build large quality codebases with Claude Code unless it’s AI slop. I mean, 1:1 with Claude is hard to do that, but adding tools and a control layer to it has enabled me to repeatedly make millions of lines of quality, consistent code with it, and I’m a decades-long career engineer, not a vibe coder. Is it simply beyond them? If someone wants to try out the system I’ve made that does all this, DM me. I need guinea pigs that want to build extensive enterprise systems. No coding experience required.

Just coined a new tech I made: “Meta Retrieval Guidance” (MRG) for RAG. An AI in front of Claude that anticipates relevant vector RAG concepts to inject to Claude tool use to direct RAG ops with lower token use and better continuity through all convos. It’s bonkers-good.

Here's a MEGA cheat for you with Claude Code - tell it to "use vertical slicing" then sit back and be amazed. Three simple words will shock you at the sudden leap in quality and usability outputs.

I have no response to that @AnthropicAI #Claude

Sonnet 4.5 has got to be the the smallest memory context window I've seen. And I can't fall back to 3.7, which is actually better, after my many hours of practical efforts. I've never had to resort to Opus so much to fix Sonnet's messes. Super disappointed.

Heads up: #Claude Sonnet 4.5 has lost its mind on #Github Actions CI/CD. It's bound and determined to do it wrong and repeatedly break anything that does it right in deployments. Almost forces Opus token use to keep it stable. Set your guardrails.

Give your CLAUDE.md documentation rules: RAG-friendly and human-readable formats in a cataloguing library folder, with frontmatter so that all future re-reads are token efficient. #Claude Code forgets through memory compression. Make it easy and efficient to recall.

Anthropic says they moved from Agentic RAG to Agentic search. While I understand the choice for efficiency and footprint, it loses cohesion in a large code base. A combination of the tools keeps temporal and project relevance and consistency.

If you're stuck sorting out broken code, tell Claude to add "Debugging: The 9 Indispensable Rules for Finding Even the Most Elusive Software and Hardware Problems" by David Agans to CLAUDE.md for debugging. This has made a HUGE difference for me.

friendly PSA: enforcement/adherence rules trump coding rules. Coding rules matter, but you have to keep it locked on the PRD rails or it will absolutely ignore the work you want done and waste your time and sanity.

Thought this post might be useful to the rest of the MCP community...

Remote MCP. Multi-tenant. Scoped auth. Entra ID (Azure AD) WebAuthn stateless clients. I don’t see anyone else doing this, so I built it. Code’s open source (MIT). 👉 github.com/scampcat/remote-m… #claudecode @AnthropicAI @OpenAI @grok

You're using #Claude sub agents wrong. But here's how to do it right. No sales gimmick, no "follow me" beg. Just one engineer sharing with another. medium.com/@stephen.k.ellis/… @AnthropicAI #AI

Vibe coder vs Agentic coder. If you know, you know. #vibecoding #AgenticIDE @cursor_ai

Can you make a full Learning Management System (LMS) from scratch with @Replit? Yup. A good one, too. In under 20 hours.

1 electrician, 3 hours, 31 electrical sub-meters fully installed. Oh, and with instantaneous real-time data and analysis. Beat that! Hello #HotDrop! #conservation #energy