The single-task harness adaptation loop is the core procedural cycle used to distill an execution error into a concrete runtime control:

1. Execution: Execute the current profile on a given task instance.
2. Observation: Observe the execution failure or external verifier feedback.
3. Attribution: Attribute the root operational cause of the failure.
4. Abstraction: Abstract a reusable procedural rule or invariant.
5. Encoding: Encode the change directly into the agent-native runbook (e.g., adding an evidence check, updating a state prompt, or altering transition guards).
6. Validation: Re-execute the updated runbook to validate that the failure is resolved without introducing regressions.

Because the control layer is agent-native and CLI-operable, the task executor and profile maintainer can be the same autonomous agent.