promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_promptdojo_

CLAUDE.md, AGENTS.md, .cursor/rules — the system prompt your agent reads every session — step 6 of 9

The current CLAUDE.md has the rule "Write maintainable code." It's vague — the model can't operationalize it. A reviewer can't mark it pass/fail. It's fluff.

Rewrite the rule as something specific the model CAN follow: "Functions over 50 lines: split. Files over 400 lines: split."

The script counts how many rules pass the "actionable" check (defined here as: contains a numeric threshold OR a forbidden/required verb like "use", "don't", "prefer").

Expected output:

actionable: 3 / 3
The break is on line 5 — but read the whole snippet first.

full-screen editor opens — close anytime to keep reading.