I built RepoMethod because coding agents are becoming capable enough to handle much larger parts of the implementation loop, but I didn't want the engineering process itself to live inside one model, chat or provider-specific prompt. RepoMethod keeps scope, acceptance criteria, verification and delivery rules inside the Git repository. The agent can implement the task and run the real project checks, while the repository decides whether the resulting state is actually deliverable. I tested it end-to-end with ChatGPT on a public TypeScript repository. A valid change passed delivery, while another change with green tests was blocked because it touched a file outside the agreed scope. I'd be especially interested in feedback from people already using coding agents for real repository work: what do you keep in the agent runtime, and what do you think should live in the repository itself?

RepoMethod

Repository-owned engineering for coding agents

Repository-owned engineering for coding agents