Schemas

Protocol version 1 uses JSON Schema Draft 2020-12:

https://json-schema.org/draft/2020-12/schema

The canonical model has exactly three pillars:

• Vision — research, prd, context
• Loop — triage, design-document, spec, ticket
• Evidence — adr, audit, record, drift

Drift is an Evidence type, not a pillar. The first nine types are Markdown document families. Records and Drift Events are JSON envelopes.

Exports

@langdrift/sdk/schema exports:

• SCHEMA_VERSION, currently the literal 1
• schemas, one schema for each artifact type
• projectSchema
• pillarSchemas.vision, .loop, and .evidence
• artifactPillars, the type-to-pillar mapping

Each pillar schema embeds the definitions and artifact shapes it needs, so the generated schema file is self-contained.

Validation behavior

Schemas require explicit schema_version: 1. Artifact IDs are nonempty and contain no whitespace. Markdown metadata requires type, schema_version, id, title, and at least one owner. Record, Drift Event, reference, actor, and project fields are defined separately.

Objects reject unknown properties where the protocol defines a closed shape. The validator does not coerce types, apply defaults, remove fields, or migrate unsupported versions. Format checks cover date, date-time, and other protocol-specific strings.

x-langdrift is protocol guidance beside standard JSON Schema assertions. It records information such as document type, version, package-owned source glob, frontmatter namespace and required fields, reference relationships, pillar, and payload resource. A general JSON Schema validator can ignore this extension; LangDrift uses it to describe interpretation.

The schemas describe portable inputs. They do not define a Product Vision formula or prove that a backend, integration, or MCP server exists.