ROADMAP.md

Package roadmap

The repository starts with the small, already exercised package set listed in catalog.json. Historical code is migration input, not automatic authority.

• @velarscript/script-analysis: restore only when an editor actively consumes the bounded local JavaScript/TypeScript analysis surface. One verified defect from the previous implementation is waiting in docs/restoration-findings.md: re-lexing with no preceding-token context, so one typed space turns the rest of a line into a regex literal and rename silently drops occurrences.
• @velarscript-labs/sqlite 0.2 is a fresh Worker-isolated implementation rather than a restoration. One connection serializes admitted work, a transaction callback owns its scoped handle, reentrant connection use fails immediately, and close rolls back through Node's real isTransaction state. It deliberately omits the old prepared-statement, streaming, migration-model, and ORM surfaces. The seven historical defects in docs/restoration-findings.md remain regression and admission evidence for any future expansion.
• @velarscript-labs/database 0.2 is the small engine-neutral layer established by the SQLite and OpenVoxel consumers: named parameterized commands, typed queries, affected-row invariants, per-operation row bounds, and a three-function executor. It is not an ORM and will not grow entity tracking, relations, inferred schemas, a query builder, or a portable migration language without separate multi-consumer evidence.
• Deployment-provider packages: admit as integrations under the same package, compatibility, and release rules; never hide them in Core build behavior.
• A standalone LSP client: extract only after two editor hosts duplicate the transport, cancellation, restart, or coordinate-conversion implementation.

Findings carried over from implementations that predate this repository are collected in docs/restoration-findings.md. They remain design constraints and regression evidence, not a promise to restore the old APIs.