Roadmap

Status: pre-release, working toward v0.1.0.

The runner works. What is missing is distribution.

Done and verified

• Hermes-native execution: sealed IIFE per test file, spawn-per-file, nonce-framed result channel, hardened against forgery and prototype pollution.
• Test discovery with globs, bounded-concurrency parallel execution.
• Jest-shaped surface: describe / test / it, the four hooks, .skip / .only / .todo, async tests, expect.extend, assertion counting.
• Matcher set: equality, truthiness, numeric, string, collection, object, toThrow, .resolves / .rejects, call and return matchers.
• Source maps — failure stacks remapped to the original TypeScript.
• React Native native mocks: NativeModules / TurboModuleRegistry shims plus a user-facing registration API. argus.fn() and argus.spyOn().
• Component testing on real React inside Hermes: render, rerender, unmount, screen, getBy*, queryBy*, within, fireEvent, act.
• Engine and version resolution from the project’s own React Native install, with the offline lookup table as fallback.
• The Hermes build-and-publish pipeline, gated on bytecode parity with the official hermes-compiler — with the first release cut: hermes-bin-v250829098.0.16, V1, all four platforms, covering React Native 0.86 and 0.87.
• 372 host-side unit tests, plus 15 Hermes fixtures — 6 passing, 2 intentionally failing, 7 adversarial.

v0.1.0 — Make it installable

The only milestone that matters right now. No new test-authoring features ship before it.

Distribution

• Build pipeline emitting JS and .d.ts for every published package.
• Replace monorepo-layout path resolution with node resolution, so the CLI finds the framework and polyfills when installed under node_modules.
• A working argus bin entry.
• Publish flow with complete package metadata: license, repository, description, files, engines, publishConfig.access.
• Cut the remaining hermes-bin-v* releases so the prebuilt step stops reporting a 404 on React Native 0.83 through 0.85 and on the legacy engine.

Engine target

• Default to V1, with V0 as an explicit opt-in target.
• Detect the project’s engine and warn loudly on mismatch.
• Assert the VM’s HBC bytecode version against the project’s pinned hermes-compiler at startup.
• Scope the class-lowering plugin to the legacy target only.

Provisioning

• Publish prebuilts for darwin-arm64, darwin-x64, linux-x64 and linux-arm64.
• Refresh the RN-to-Hermes lookup table in CI from the published branches table.
• Bundled-VM detection for the legacy target on RN 0.73–0.82.

Configuration

There is no config file today. The design is settled:

--config <path>
argus.config.ts
argus.config.mts
argus.config.js
argus.config.mjs
.config/argus.config.ts
package.json → "argus"
built-in defaults

Precedence, lowest to highest: defaults → package.json field → config file → environment → CLI flags. Loading uses native import(); Node type-strips argus.config.ts directly, so this needs no dependency and no bundler.

• ArgusConfig type and documented defaults.
• Config loader and merge layer, wired through discovery, bundling and provisioning.
• include / exclude replacing the hardcoded globs.
• passWithNoTests — a zero-match run currently exits 2.

Make component testing optional

Testing plain TypeScript should not require the component facade. The runtime already behaves correctly; what remains is packaging plus one genuine design fix — React is currently located relative to the @argus/rntl install directory, which makes it structurally mandatory for any JSX.

• Optional componentPath in the bundler input and the CLI path resolver.
• Resolve React from the user’s project, independent of @argus/rntl.
• Conditional alias map, with a clear diagnostic when argus is imported without the package installed.
• Regression test: bundle a pure-TypeScript suite with no React installed.

CI

• Build on CI; run the Hermes fixtures there, not just locally.
• macOS and Linux matrix.
• Release workflow.

v0.2.0 — Test-authoring features

Deferred until v0.1.0 ships. Valuable, but none of them block adoption.

• Snapshots — toMatchSnapshot(), external .snap files, --update, obsolete detection and safe pruning. Design work is complete.
• Coverage — provider selection, thresholds, reporters.
• Watch mode — file watching, re-run on change, interactive filtering. Worth splitting compile from execute here: hermesc -emit-binary to a cached .hbc, then hvm to run it. Measured at roughly four times faster on re-runs, because compilation dominates and the cache absorbs it.

v0.3.0 and beyond

• Reporters — pluggable output, JSON and JUnit for CI.
• CLI framework migration — worth doing once a second subcommand exists (init, watch). Until then the built-in parser is adequate.
• argus init — scaffold a config file.
• Windows support — needs a prebuilt VM for win32-x64 or a documented source-build path.
• Broader React Native range — validate and document which versions work.
• Setup files — user-supplied global setup and teardown.
• bail / fail-fast, retry, test-name filtering.

Deliberate non-goals

Running on a device, full Jest compatibility, replacing Vitest for non-RN projects, and bundling a JavaScript engine other than Hermes. Reasoning: Limitations & non-goals.