Every project has its own way of doing things. Today every agent run already reads the repository's AGENTS.md and CLAUDE.md, at the root and in the directories it touches, and reviews read .g1t/review.md, all from the default branch. Playbooks build on those files with structure g1t can act on, not just read.

What it will do

• Instructions per kind of work: fixes, features, reviews, upgrades
• Commands to build and test that g1t runs before every change, not just tells the agent about
• Paths agents may not touch without a person, enforced on push
• Learned from reviews: what people corrected becomes a proposed rule