ralph

To enforce a Definition of Done, you need an iterative task-completion loop that derives concrete completion criteria up front and runs objective checks like tests or lint commands before allowing exit. This prevents incomplete work by blocking session end until all criteria pass.

Iterative agent verification is a process that validates code changes against an agreed Definition of Done using a dedicated verifier agent. It objectively marks checklist items as pass or fail based on actual files and tests, reinjecting the prompt with remaining items if checks fail.

You can stop an agent from exiting prematurely by using a stop-hook mechanism that blocks session end when any Definition of Done items remain unchecked. It reinjects the original prompt with the remaining checklist items to trigger targeted fixes.

Yes, a DoD-first workflow turns your bug fix intent into three to seven binary, verifiable completion criteria before any work starts. Interactive confirmation ensures you agree on these checks, such as expected files or no type errors, prior to execution.

Iterative verification runs without external dependencies, relying on a persistent DoD file and session state to track checklist progress. It applies to coding changes where completion must be proven through objective checks like build commands and file existence.

To avoid endless cycling during verification, the process uses a max-iteration circuit breaker. If the loop reaches its maximum iterations without satisfying all Definition of Done items, it stops to prevent infinite cycling.