It adds an explicit operating layer around the agent by modeling product context, generating strategy versions, producing prioritized tasks, enforcing approval gates, executing work, and evaluating outcomes through controlled state transitions. Workflows are declared in YAML and all entities such as tasks, artifacts, evaluations, and strategy revisions are persisted in SQLite, making the system inspectable and replayable.

Each agent role follows a defined rubric, and execution results feed into a scheduled strategy evolution loop that reassesses context and prior performance before proposing new work. Nothing runs without approval, but iteration continues autonomously based on observed outcomes and stored state.

The goal is to explore whether growth can be delegated to an agent operating inside a constrained feedback system, where adaptation emerges from measurable results and decisions are governed by explicit rules rather than informal adjustments.