Agent-Native Repo Playbook

Outcome

Make the repository a reliable harness in which a cold agent can take clear intent through implementation, validation, product-facing proof, cleanup, and reporting with minimal human coordination.

Operating model:

• Humans own intent, priorities, acceptance criteria, taste, and material risk.
• Agents own code, tests, docs, tooling, validation, cleanup, and routine follow-through.
• Repeated confusion or failure is a harness gap to fix, not a prompt to repeat.

Use recommendation-only mode for audit, review, score, or report requests. Edit the repository only when the user asks for changes or implementation.

Modes and References