Engram Plan

Use this skill when turning the engram architecture into implementation slices. Plans should stay modular, composable, and contract-backed: Rust owns the deterministic core, TypeScript owns integration ergonomics, and adapters sit behind stable ports.

Required Context

Read these files before planning implementation:

• docs/domain-data-model.md
• docs/architecture.md
• docs/research/synthesis.md
• docs/rfcs/0001-memory-layer-scope.md
• docs/rfcs/0002-knowledge-source-extension.md
• references/crate-map.md

Planning Rules

• Treat the accepted domain model as the source of truth.
• Keep SQL, vector stores, model providers, and Node runtimes outside the Rust core domain crate.
• Prefer ports and adapters over baked-in infrastructure.
• Plan vertical slices that can be tested with in-memory adapters before persistent storage.
• Make evaluation part of each milestone instead of a separate late-stage project.
• Call out contract risks before implementation risks.

Default Build Order

1. Contract freeze review and implementation stack ADR.
2. Rust domain crate with serde-compatible types and invariants.
3. Rust engine crate with ports, in-memory store, policy gate, and retrieval pipeline skeleton.
4. Evaluation fixtures and deterministic regression harness.
5. Adapter crates for SQL/vector/object storage.
6. N-API TypeScript package and generated TypeScript contract types.
7. Ingestion extensions for code and unstructured documents.
8. Background consolidation, belief derivation, hierarchy maintenance, and observability.

Output

Plans should include:

• Crates and packages affected.
• Contract surfaces involved.
• First testable slice.
• Explicit deferrals.
• Verification commands.