Hi HN! I wanted to test the custom skills I've been building for Claude (https://github.com/josh-gree/gen-art-framework – see PRs/issues for the workflow patterns), so I asked it to build an automated generative art gallery from scratch.

  Result: https://josh-gree.github.io/gen-art-gallery/
  Repo: https://github.com/josh-gree/gen-art-gallery

  Everything here was done through conversation with Claude:
  - Set up the repo structure and GitHub Actions workflow
  - Built the static gallery site (HTML/CSS/JS)
  - Debugged workflow failures (Python version issues, uv configuration)
  - Added incremental generation (only regenerate changed scripts)
  - Added UI features (modal view, source links)
  - Created example art scripts

  The skills handle the full development lifecycle:
  - Creating tickets to capture intent (not implementation plans)
  - Planning implementations for existing tickets
  - Executing on plans
  - Reviewing PRs and addressing feedback
  - Managing git worktrees for parallel development

  The goal wasn't the gallery itself, but pinning down a good set of patterns for how Claude should handle planning, implementing, and reviewing changes in a real project. The PRs and issues in the gen-art-framework repo show the iteration on these workflows.

  Curious if others are building similar "meta" workflows for AI coding assistants – patterns that work well across different projects?