AGENTS.md, CLAUDE.md, WRITING.md target AI slop

The useful part not novelty in file naming, it’s discipline: evidence before edits, narrow changes, and no “done” claims without checks.

• –The coding rules directly attack the most common agent failure mode: jumping into code before the repo is understood and then sprawling into adjacent files
• –The emphasis on real verification and explicit parallel work is especially relevant for open-weight models, which usually need more structure to stay grounded
• –WRITING.md is a good mirror image of the coding rules: it treats prose like an engineering artifact, with clear bans on filler, repetitive cadence, and over-polished vagueness
• –These repos are most valuable as defaults you can adapt, not as a universal doctrine; the best parts are the concrete failure cases they encode