← ClaudeAtlas

feynman-findingslisted

The one way code review findings are written for this user, used by /judge-code and /review-code. Each finding is a concrete example first, then how bad it is, when it hits and what happens if we wait, then the fix and its effort, and the file name last. Also holds the four Feynman tests reviewers run on the code. Load it whenever you report review findings, audit results, or a list of problems.
whoiscarlo/agent-skills · ★ 1 · Data & Documents · score 64
Install: claude install-skill whoiscarlo/agent-skills
# Feynman Findings The user asked for this on 2026-11-13: "break down the issues into simple concrete examples... and what the fix you are suggesting... Too many words confuse me." A first version had "pro and con" lines; the user said they added nothing and asked instead for what shows "how severe the issue is" and whether "it needs to be tackled now or not". It draws on the installed `feynman-technique` and `feynman-principles` skills. Feynman's order for explaining anything: **something you can do → what you would see → only then the name.** A finding follows the same order. The file and function name come last, because a name explains nothing. ## Part 1: the four tests reviewers run on the code Give these to every reviewer, next to their other instructions (from `feynman-principles`, turned toward code): 1. **Say it plainly.** Can you say what the code does in everyday words, with no name doing the work? "It handles retries" fails. "When a search fails, it waits an hour and tries once more" passes. A step you can't say plainly is a step nobody checked. 2. **How would we know it's wrong?** For each behavior, what would we see if it broke: a log line, a row in a table, a number that stops moving? No answer means no test and no alarm. Flag it. 3. **Take it out.** Remove a piece in your head. If nothing changes, it's decoration: say so. If something breaks, follow it: what breaks next, and next? 4. **Look at the mess.** Run the bad days through the code: the provider is