WebSeitz/wiki

(2025-11-04) Claude web, android, code, skills, and directories

I've read multiple pieces from people recommending Claude Code for non-coding tasks. And it smells like Claude Skills could also add value. But I'm confused as to

  • when/how/where do I create local project-specific directories? How should such directories be organized? Does that change if a project will involve both coding and non-coding tasks/docs?
  • once I start a project in Claude Code, do I always have to work in the Terminal of my same machine? I can't use Claude-web or Claude-android to move the (non-coding) work along?

Summary

(scraped/processed a bunch of articles I've read in the last couple days - see backlinks at the bottom of Claude Code and Claude Skills)

Also, see

I think I can work with my current folder structure

  • all my code is in ~/code/ (subdirectories by technology)
  • all my indie-thinking/writing is in ~/fluxent/
  • each dayjob gets a subdirectory of ~/career/

I only need #2 for now

  • I don't think I'm going to have separate Projects yet

Nov06

  • cd ~/fluxent/
  • claude code
  • /init to make claude.md -> This directory does not appear to be a software development codebase... Creating a CLAUDE.md file would not be appropriate here
  • also, I have a ~/.claude/ but it doesn't have a claude.md in it, either
    • it has projects/-Users-billseitz-Documents-fluxent/ directory, but no claude.md in there, either
  • well, Teresa's article (link above) shows parts of her files, so clearly there's no scary format to worry about. So I'll try just making a couple by hand, and then restarting CC and see if it complains.
  • use touch claude.md in each folder to create empty file, open in text editor, add some markdown
  • quit/relaunch Claude, no comments about the files, moving ahead
  • I have a CSV dump (from Oct09) of my PrivateWiki, so I want Claude to ingest it... The goal is to help Claude understand Bill's thinking patterns and knowledge base. So analyze/summarize/save to a format that will be efficient for your re-use.
    • tool uses · 76.9k tokens · 2m 55s... Wrote 396 lines to BillSeitz_Context.md
  • do the same for the 127 WebSeitzWiki links listed in a txt file, all related to Agile Product Development: Now do the same thing, by reading all the links in file agileprod.txt and saving the result to file AgileProductDevelopment_Context.md.
    • hmm is it getting econnrefused?
    • many steps of text processing I don't understand. In retrospect I should have done a giant query to export the 150 pages in the save CSV format as for the private....
    • finally done, total cost 160k tokens
  • figure out what to do with the skills files for writing-viral-tweets I got from Tom Crawshaw
    • ah actually they're not skills files, they're just "context" files, so I'm just going to move them into a subfolder and....?
    • Teresa article says For more detailed context, create separate context files. You can then reference these as you need them. You can even describe in your CLAUDE.md files what context files exist, and Claude will automatically know to use them.
    • so made a md file at top of /fluxent/, referring to all the subfiles using @ path reference.
    • (also added references to these md files in claude.md).
    • then restarted CC
  • get warning about Large context, "will impact performance". I see varying statements online about memory/performance vs CC "deciding when to use" specific files. Does the @ path matter?
  • using it now: Using all that context, generate 40 posts of max 256 characters each, each reflecting my opinions on Agile Product Development. Make roughly half of them negative/critical of bad practices and the rest of them positive, encouraging good practices. Save them to AgileProdDevToots.md

Nov06 reducing context

  • tell CC You can drop @x.md from this session's context. It replies: Understood. I won't reference the Tom Crawshaw writing guide materials for the rest of this session.
  • ask Does this improve memory/performance by reducing context?
  • Answer: Not really, no. Those files were already loaded into the session context at the beginning (via the system reminders you saw). Just agreeing not to reference them doesn't remove them from the conversation history - they're still consuming tokens in my context window.
    To actually reduce context usage, you'd need to either:
    • 1. Start a new session without those files loaded
    • 2. Remove or comment out those references in your CLAUDE.md files before starting a session
  • The system reminder earlier showed token usage: 54,936 tokens used out of 200,000 budget. A significant portion of that came from loading those large context files (especially the AgileProductDevelopment_Context.md which was ~1,600 lines).
    If you want to optimize for performance/memory in future sessions and don't need those specific guides, you could temporarily comment them out in your CLAUDE.md configuration.
  • I asked it how to make it more dynamic: Here's the best approach: Create a Lightweight Catalog: Instead of including the full files in CLAUDE.md, create a brief index/catalog that describes what's available and when to use it. Then I can use the Read tool to load files only when actually needed.
  • How This Works
    • 1. I see the catalog and know these resources exist
    • 2. When a task requires specific knowledge (e.g., "write a blog post about SAFe"), I can then use Read to load only the relevant file
    • 3. Context only consumed when actually needed
  • and then it offered to fix my claude.md and I accepted

Edited:    |       |    Search Twitter for discussion

Made with flux.garden