How I Build

Six steps, every project.

1. Start from a real problem. Typing the same work order template on every call. Camera apps that buzz for every passing car.
2. Write the plan first. A spec says what the app does and what it doesn't. For Sightglass, each feature is its own GitHub issue.
3. See it before it's built. Claude mocks up two or three designs, and I pick one or mix the best parts.
4. Build on a branch. Work happens on a side copy in small saved steps. The live version isn't touched until I approve.
5. Check the work. Automated tests, speed, accessibility and security checks, then testing on the real thing: my iPhone, my cameras, my TV.
6. Write it up. Each project keeps a build log of what broke and what I learned, and a handoff note so the next session picks up where the last one stopped.

Example: this website.

1. Five rounds of mockups, with three options each time, until the look was right.
2. The final look started as a screenshot I saved. Claude matched it with free stock photos, and cut the pipes and cameras out of their photos so the big titles sit behind them.
3. Checked before going live: Lighthouse scores of 99 to 100, no accessibility errors, strict security headers, and no personal info in any photo.
4. Went live with one merge, the same day the build was finished.

The rules I give Claude.

One rules file, synced to every computer I use, that every Claude session reads before it starts. A few of them:

• Work on a branch, never on the live version.
• Never save passwords, keys or customer info in the code.
• Ask before anything that costs money or can't be undone.
• Save progress before stopping, so a crash never loses work.