Phase 3A: Implement Agent Manifest schema and validation

[yacosta738]

Parent epic: DALLAY-249
Source of truth: tmp/PRD-Capability-Based-Architecture-for-Composable-AI-Agents.md

Phase 3A: Implement Agent Manifest schema and validation

Objective

Define the TOML manifest schema for composed agents and validate manifests before composition.

Scope

• implement the Agent Manifest schema described in the PRD
• support core sections such as agent, providers, channels, tools, memory, observability, security, loop, and identity
• implement validation rules including:
  • at least one provider
  • at least one channel
  • default provider must be enabled
  • enabled capabilities must be known
  • memory backend must be valid
  • security sandbox must be valid for the target platform
  • tool restrictions must be subsets of supported restrictions
• reject inline secrets and prefer env-var references where applicable

Acceptance Criteria

• valid manifests parse successfully from TOML
• invalid manifests fail with useful validation messages
• validation rules are covered by targeted tests
• schema naming matches the PRD and future composer expectations

Notes

This issue defines the declarative contract that later composition flows depend on.

[linear]

DALLAY-261