← sources

Non-dev here, about 1.8M lines of Claude Code output later. How do you audit code you can't read?

source: reddit (reddit:ClaudeCode)published 2026-11-11 05:56ingested 2026-11-12 02:42

retained text

I don't have a dev background. I've been building with coding agents almost every day since late 2024, Cursor first and now mostly Claude Code, all from plain English. My projects are around 1.8 million lines of code by now, and I still have the transcripts of a few hundred of those sessions.



My guess is that people like us need code review more than developers do, not less. A developer usually writes a tighter spec and can read the diff. When I ask for a feature, a lot of what I never spelled out gets decided by the agent, and I can't read the code well enough to see what it chose.



So I want to know how other non-programmers handle this. Do you run any kind of review or audit after a session? Which tool or prompt? Can you actually follow what it tells you? And what happens to the findings, do you paste them back to Claude, fix some, or let them go?



I'm going to run the tools you mention on my own code and sessions and post what each one found, and whether someone without a dev background could understand it and act on it. What I have so far: Anthropic's code-review plugin, /security-review, CodeRabbit, and a plain prompt asking Claude to explain the changes in plain English. Tell me what you use and I'll add it.



Hoping this ends up as a review routine that works for people who can't read the code.

document fetched 2026-11-12 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xta125/nondev_here_about_18m_lines_of_claude_code_output