Feat — execution specifications

Write the spec. Predict every effect. Anything unpredicted is a failure.

.feat files are execution specifications: they instruct how a feature is built and predict its complete observable footprint — the response, every database write, every file touched, every event published. They compile deterministically into complete test suites with zero human-authored test code. This extension makes the spec a live document:

• Syntax highlighting for the full .feat language — identity, triggers, predictions, matchers, and stimulus references each carry distinct scopes, so the spec reads as structure, not a wall of keywords.
• Live spec-anchored feedback: the extension runs feat watch --json behind the scenes. When an implementation violates a prediction, the failing scenario line squiggles in the spec itself — not in a generated test file. The status bar shows green or the current violation count.

Setup

1. Install this extension.
2. Open a project containing a feat.config.json with @mmmnt/feature installed.

That's the whole setup. Watch starts automatically (disable via feat.watch.autoStart); Feat: Start watch / Feat: Stop watch control it manually.

The language in one glance

scenario "successful user creation":
  when: CreateUser { email: "alice@example.com", name: "Alice" }
  predict success:
    response 201 UserResponse { id: any uuid, email: @when.email }
    database has [ INSERT with UserRow { email: @when.email, status: "active" } ]

Prediction inversion is the contract: a predicted effect that doesn't happen fails, and an effect that happens without being predicted also fails. The suite that enforces it is generated, byte-deterministic, and locked to the spec by feat verify.

Part of Feature — docs at the project wiki.