Claude Code - awesome but needs a great setup, what is yours?
by•
Claude Code is amazing. Agentic workflows just... work - at least better than last years options for vibe coding.
Things I've learned:
- .md files are a must
- architecture and context
- log history after every pass
- Short passes with initial creation, full pass with design is fine
- Fable is a hog.
- use ChatGPT or Gemini for UI/UX and design prompting, but not before you get Claude to chop down your architecture first to hand off.
- /clear
What other things do you use to ensure Claude is productive and doesn't break with a long term project?
15 views
Replies
My Claude Code setup starts with choosing the right skills and MCPs for the specific project. I try not to overload it with every integration available. I only give it the tools, documentation, and context it actually needs.
I also define the architecture, coding standards, and project goals up front, then break the work into smaller phases with a verification step after each one. That keeps Claude focused and makes it easier to catch issues before they spread across the codebase.
For long term projects, the quality of the context and tool setup matters almost as much as the model itself.
my CLAUDE.md isn't just architecture notes, i keep a running log of stuff that already went wrong so it doesn't get quietly rebuilt again in 3 months. built a whole caching layer once, it got patched twice then fully deleted because the shape of it was wrong from the start. that whole saga is like 3 sentences in the file now instead of a repeat incident
also started writing down false positives, not just facts. almost had two files deleted as "dead code" because the search only checked src/ and missed root level files importing them. that kind of thing never occurs to you to document until it bites you once
As of right now, my workflow relies heavily on 2-3 LLM's
Sol 5.6 to take my long winded ideas or comments to translate them to PRD
Running Task list within the PRD that is updated upon each pass.
GPT Sol 5.6 High to organize context and meta prompt to Claude code back and forth. Establish packets for each implementation
Let Sol 5.6 tell me when to clear context, what Claude model to use and what effort based on the implementation difficulty occuring within the prompt so i can use my remaining brain power on distribution while i build.
Gemini to do packet reviews post implementation, ChatGPT Sol creates a packet check list to send to claude to create a zip file of the code included of the pass, which is uploaded to both ChatGPT and Gemini. If they are both aligned, I move on.
Im gonna begin live streaming this upcoming week as I build to launch my product in September!