Status and Roadmap

Status and Roadmap

Scope of record: FEATURES.md in the repository — every capability is a status-tracked row.

Built and self-hosting (today)

• The language, v1 — dual zones, four closed reference spaces, matchers (references, typed wildcards, regex, absence, nesting, golden fixtures), seeds, event triggers (deliver), state assertions (contains), the query guarantee, actors, clock freeze, scenario outlines, cross-spec dependencies, the agent contract (lifecycle, touches, rejects traceability).
• The corpus — nine exemplars with expected-IR goldens; the parser's conformance suite.
• The toolchain — parse / generate / verify / run / report / audit / lint / fmt / watch / diff; deterministic emission; JUnit output.
• Adapters — handler (direct invocation), http, fs (filesystem capture).
• Editor — TextMate grammar (editor/feat.tmLanguage.json).
• The repository itself runs on the chain it ships: feat verify && feat run (all suites generated from .feat specs).

In front of us

• Release engineering — CI pipeline for the repository (verify + run + workspace tests as required checks), versioned releases, npm publishing.
• Quality gates — coverage thresholds, adapter compliance suite (feat-adapter-kit), config JSON Schema published at a stable URL.

Deferred (designed, not yet built)

• Runtime wiring for contains and deliver — the grammar, IR, and derivation are complete; execution needs event-store service adapters that can deliver stimuli and serve reads.
• External service adapters — EventStoreDB, DynamoDB (+ Streams), Kafka, LocalStack, gRPC response adapter, TypeScript schema adapter.
• LSP / editor depth — a language server (diagnostics, completion for the closed spaces, go-to-definition for schema references); the corpus keeps any second grammar honest.
• feat import — bridges from other spec formats. Imports land as status: draft with a gap report ("what your spec fails to measure") and must clear the full quality gate before anything is built — on-ramps, never bypasses.
• Agent authoring surface — an MCP server exposing parse/lint/audit/gap analysis to AI assistants for collaborative spec authoring.
• Adapter ecosystem kit — create-adapter scaffolding + compliance test suite for third-party adapter authors.
• External validation — running Feature against a production event-sourced system.

Design principles that won't move

1. The language is the product; runners and adapters are ecosystem.
2. Predict complete; reject everything else.
3. Zero glue code — if predictions can't express it, fix the language, never add an escape hatch.
4. Agent zone freeform, compiler zone strict.
5. Generated tests are committed, self-contained artifacts.
6. Deterministic everything: same inputs, same bytes.
7. A language change is incomplete until the reference, a corpus exemplar, the implementation, and full corpus validation land together.