I'll take a stab at defining high level set of principles around what Agents should be able to do:

Agents should:

• Be able to react to events and raise new events
• Be able to interact with other agents (Ask - expects a response, Tell - only ack.)
• Have state and memory
• Have specialisation (ex. developer, architect), capabilities (ex. implement code, run code, test code), knowledge (WAF framework) and context (understanding of the repo)
• Be able to be part of a cluster (group)
• Have unique identity, ideally hierarchical (scoped) (ex. Org dev lead, Repo dev lead, Intent dev lead)
• can ask another agents for their memory recollection instead of it having access to hierarchical memory