The Skill Soup: September Recipes That Worked

Instead of theorizing about combined skills, here are the actual combinations I used on client work this month, ranked by how well they worked. Concrete over abstract.

Recipe 1: Django + Claude Code + Pytest

Extending a legacy Django app. Claude Code reads the existing models and views, generates new endpoints in the same style, and pytest catches regressions. Time to first green test went from a day to forty minutes. This is the combo I keep reaching for.

Recipe 2: Scrape + Whisper + Markdown

Turning hours of Zoom recordings into a searchable knowledge base. Python scrapes the video URLs, Whisper transcribes, a small script chunks the output into markdown files. Three tools, zero proprietary software.

for video in videos:
    audio = download_audio(video.url)
    text = whisper.transcribe(audio)
    write_markdown(video.slug, text)

Recipe 3: Next.js + Strapi + Vercel

A client blog that editors can own. Strapi for content, Next.js for display, Vercel for deploys. The client doesn't need me to publish a post. This is what success looks like: the system runs without me.

Recipe 4: Claude Code + WordPress + rsync

Building custom WordPress plugins with Claude Code in a local environment, then rsync to staging. Fast iteration, no CI. Fits the scale of the work.

The One That Failed

I tried Cursor + remote SSH + a Docker container for a client's legacy PHP app. The latency and the layer boundaries killed every workflow. I switched to VS Code with direct SSH and it was fine. Lesson: if the latency feels wrong, stop. Trust the friction signal.