AI Agent Contracts Guide Code Edits But Cannot Enforce Repository Acceptance

AI agent contracts are effective at governing how an agent should modify a repository — covering rules around sensitive files, diff size, and handoff expectations — but they cannot define whether a change has been formally accepted. Acceptance is an execution question involving setup paths, dependency installation, and test verification, which falls outside the scope of contribution rules alone. A tool called Ota addresses this gap through a machine-readable configuration file, ota.yaml, that explicitly declares safe tasks, setup sequences, and verification steps for agents to discover and execute. Ota enforces an agent-admitted execution path via its runner boundary, ensuring agents only run pre-declared, safe task closures rather than inferring steps on their own. The recommended model keeps agent contracts responsible for editing behavior while Ota governs repository acceptance, with the two working in tandem rather than as substitutes.