Typed artifact schemas per layer/transform

[marklubin]

Source

From SYNIX_REFLECTION_AND_BACKLOG.md — P2 #10.

Problem

App builders repeatedly parse generic JSON artifacts. Every downstream consumer such as UI, API, export, or runtime glue writes custom parsing logic because there is no schema contract per layer or transform type.

With built-in chunk, summary, core-memory, graph, and runtime tool payloads, this gap becomes more expensive. Without explicit schemas, every consumer must reverse-engineer platform outputs.

Recommendation

1. Optional typed schema metadata per layer or transform, including built-in artifact families and standard tool payloads
2. Schema declaration in layer or built-in config
3. Optional validation that produced artifacts conform to declared schema
4. Documentation that exposes those schemas as part of the platform contract

Acceptance Criteria

• UIs, APIs, and runtimes can render or consume layer-specific artifacts with less custom glue
• Schema is optional so existing pipelines work unchanged
• At least one built-in transform or artifact family declares its output schema
• Runtime tool payloads for standard bank operations can validate against declared schemas

Required Automated E2E Test

Run a pipeline that emits at least one built-in artifact family and a runtime tool response, then verify both the artifact payload and the runtime payload validate against their declared schemas.

Docs Updates Required

• schema reference for built-in families and standard tool payloads
• guidance for optional schema declaration in custom layers or transforms

Demo Or Template Follow-On

Record follow-on updates to templates so they cite and consume declared schemas instead of relying on ad hoc JSON parsing.