Type: url
Original: https://qiita.com/nogataka/items/d1b3fcf355c630cd7fc8
Status: generated
Language: ja
Summary
A Qiita article presenting harness engineering as the paradigm that follows CLAUDE.md for controlling AI coding agents: rules alone are only requests, so quality has to be enforced by structure. It traces a three-stage evolution (CLAUDE.md, AGENTS.md with split rules, harness), breaks a harness into five components (rules, skills, hooks, memory, feedback loop), reads the everything-claude-code repository as a full implementation, and recommends adding components incrementally as pain appears, escalating a rule one level each time the same violation recurs three times.