Development workflow with Claude superpowers
I have hardly written any code manually in the last year, yet I’ve built heaps of smaller and larger projects, using AI to learn new things, write boilerplate code, resolve tricky bugs, and rapidly iterate on ideas.
I have journeyed from AI autocomplete to vibe coding to prompt engineering to increasingly effective agentic spec-driven development workflows. For a while, I’ve been using and developing my trusty old prompts from my cursor-rules repo to help me keep the AI on track.
If you’re not familiar with this approach, it feels like an ironic comeback of waterfall in a way: The more time you spend planning your solution and working out requirements, and the better you are at documenting what you want accurately and effectively, the better agents are at writing code. Obviously this was true in the old era (like, a year ago) of agile development teams as well, but then, good human developers would often make up for lack of planning and sloppy specs by understanding your intentions rather than building what you asked for.
Spec-driven AI-first development has been very effective, and my old trusty prompts helped me with an organised development workflow that starts with a brainstorm session to work out the specs, then another brainstorm session to work out a technical design that meets the specs. Another prompt turned these into an iterative implementation plan, and finally the “execute” prompt would go through the plan and build it.
The prompts forced the AI to ask me one question at a time during brainstorming, do test-driven development for all coding work, and usually the results were fairly good. Other times it would go completely off the rails and either get stuck or build something completely different from what I had in mind, and almost always this was really a prompting/planning/designing issue.
@ears.mdc] B --> C[Design session
@design.mdc] C --> D[Implementation Plan session
@plan.mdc] D --> E[Execution session
@execute.mdc] E --> F[PR, Code Review & Merge]
Most effective approaches for AI-first development that I see online seem to follow a similar pattern. Recently, I switched from Cursor to Claude Code, and finally retired my own prompts in favour of Jesse Vincent’s Claude Superpowers, which has a familiar workflow but makes full use of Claude Code Skills and is generally more comprehensive and better thought out than my approach.
Check out his superpowers github repo.