Document the five-layer cognitive agent architecture

[pentaxis93]

Summary

Update agentd's documentation to describe its role in the five-layer cognitive agent architecture. agentd is the daemon layer — it orchestrates agent lifecycles. The cognitive runtime (runa) and methodology plugins (groundwork) are separate layers that run inside agentd containers.

Attachments: Architecture_Five_Layer_Model.md

Scope

• ARCHITECTURE.md — add a section describing the five-layer model and agentd's position in it
• README.md — brief mention of runa as the cognitive runtime layer

Acceptance Criteria

• ARCHITECTURE.md includes a section describing the five-layer architecture (daemon → runtime → library → spec → plugin)
• The section explains that agentd is the daemon layer and delegates cognitive pipeline work to runa
• Links to pentaxis93/runa and pentaxis93/groundwork are included
• README.md mentions runa as the cognitive runtime that agents use inside agentd containers
• agentd's existing architecture content (seven capability needs, crate layout, session lifecycle, container isolation, credential flow) is preserved — the five-layer section is additive

Dependencies

None — agentd's existing architecture is correct and unchanged. This is additive documentation.

Architecture_Five_Layer_Model.md