Agent core principles

[kostapetan]

I'd like to define a minimal, high level set of principles around what Agents should be able to do.
Here is an initial proposal:

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 multi-layered 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)

Possibly the agent can ask another agent for it's memory recollection instead of it having access to hierarchical memory?

We can have a convention in how we construct the agent IDs, so it's hierarchical in nature. We can use / to designate level, similar to how the web constructs paths.

Example would be:
agent with ID - microsoft/aigents/3, being an agent scoped at the discussions level.
When that agent wants to ask a higher scoped agent for information, it can just point to an agent with ID - microsoft/aigents or even further microsoft.