Nove Test

Continuous testing intelligence layers for AI coding agents.

Stop naive testing — simply running tests and getting pass/fail results.

Add testing intelligence layers on top of your native test engines: structured guidance through coverage-guided testing, regression testing, and fault localization.

Nove Test ships with native test runner wrappers, test result memory, a coverage analyzer, a regression engine, fault localization, and test replay — all under one CLI, all under one JSON contract.

Install · Quick start · Why · For AI agents · Docs · License

---

Install

One line, Linux and macOS:

curl -fsSL https://raw.githubusercontent.com/Nove-Lab/Nove-Test/main/scripts/install.sh | sh

One line, Windows (PowerShell 5.1+):

irm https://raw.githubusercontent.com/Nove-Lab/Nove-Test/main/scripts/install.ps1 | iex

Both scripts detect your OS/arch, download a signed binary from the latest GitHub Release, verify SHA-256, and install to ~/.local/bin/novetest (Linux/macOS) or %USERPROFILE%\.local\bin\novetest.exe (Windows). Re-running upgrades in place.

Inspect-first (recommended):

# Linux / macOS
curl -fsSL -o install.sh https://raw.githubusercontent.com/Nove-Lab/Nove-Test/main/scripts/install.sh
less install.sh   # read it
sh install.sh

# Windows
irm https://raw.githubusercontent.com/Nove-Lab/Nove-Test/main/scripts/install.ps1 -OutFile install.ps1
Get-Content install.ps1   # read it
.\install.ps1

Or download a binary directly:

curl -fsSL -o novetest \
  https://github.com/Nove-Lab/Nove-Test/releases/latest/download/novetest-linux-x86_64
curl -fsSL -o novetest.sha256 \
  https://github.com/Nove-Lab/Nove-Test/releases/latest/download/novetest-linux-x86_64.sha256
sha256sum -c novetest.sha256
chmod +x novetest && mv novetest ~/.local/bin/

For Python tooling users:

uv tool install novetest    # via uv (Astral)
pipx install novetest       # via pipx

Quick start

cd your-project          # any project — Python, JS, Go, Java, .NET, Rust
novetest init            # creates .novetest/, detects the test engine
novetest test            # runs your tests, stores the result, prints JSON
novetest status          # list stored runs
novetest inspect <id>    # full Run Record (passes, failures, coverage, metadata)
novetest compare A B     # what changed between runs A and B
novetest localization run --formula ochiai   # rank likely buggy files (SBFL)

Every command emits a novetest/v1 JSON envelope on stdout:

{
  "version": "novetest/v1",
  "verb": "test",
  "exit_code": 0,
  "data": {
    "run_id": "2026-06-10T07:23:14Z-a1b2c3",
    "summary": { "total": 42, "passed": 41, "failed": 1, "skipped": 0 },
    "results": [
      {
        "test_id": "tests/test_auth.py::test_login_succeeds_with_valid_credentials",
        "outcome": "passed",
        "duration_ms": 12
      }
    ]
  },
  "warnings": [],
  "errors": []
}

The JSON shape is stable across all six engines (pytest, jest, JUnit, go test, dotnet, cargo). Your scripts and AI agents see the same envelope regardless of what's underneath.

Why Nove Test?

You are	Nove Test gives you
An AI coding agent	Stable novetest/v1 envelopes — parse JSON, not stderr. Run Records as ground truth — compare your edits against a known baseline. SBFL fault localization — get a ranked list of probable culprit files.
A solo developer	One binary, six engines, zero per-engine config. Persistent run history without setting up a database.
A team CI integrator	Stable JSON contract across pytest / jest / JUnit / go test / dotnet / cargo. Deterministic regression detection.
A test-engine author	Documented, minimal adapter contract. Bring your own engine, get storage and cross-run analysis for free.

Features

• ✅ Adapters for six native test engines — pytest · jest · JUnit (Maven, Gradle) · go test · dotnet (xUnit / NUnit / MSTest) · cargo
• ✅ Parsers for five coverage formats — coverage.py · istanbul (LCOV) · jacoco · cobertura · with cross-drive normalization on Windows
• ✅ Persistent run records — every execution stored as immutable JSON under .novetest/
• ✅ Cross-run regression — deterministic deltas between any two runs
• ✅ Fault localization — SBFL (Ochiai, DStar, Op2, Tarantula) across three degradation modes
• ✅ Replay — re-execute a prior run under reconstructed conditions
• ✅ AI-friendly by default — --output json is the default contract; NOVETEST_OUTPUT=json env override
• ✅ Single binary — PyApp-wrapped, no Python toolchain required on the user's machine

For AI coding agents

If you're an AI agent invoking Nove Test on a user's project, here's the contract you can rely on:

• One verb, one envelope, one exit code. Every CLI verb emits { version: "novetest/v1", verb, exit_code, data, warnings, errors }. Parse JSON; ignore stderr.
• The version field is the schema contract. When it ticks to novetest/v2, the schema may break — until then it won't.
• Run IDs are stable references. Use them in subsequent calls (inspect <id>, compare <A> <B>, localization run --baseline <id>).
• Determinism is a first-class goal. A given input project + engine + version produces the same summary block. Differences across runs are user-visible flake, not framework noise.
• warnings[] is your degradation signal. Adapter quirks, partial coverage, missing tooling — all surface here as structured { code, message, ... } entries. Match on code for programmatic handling.

See docs/ for the full envelope reference and integration patterns.

Status

Production-ready for Linux, macOS, and Windows. The current release is the latest tag on GitHub Releases; novetest --version reports the build you actually have.

Stable today:

• All six native test engines and CLI verbs
• The novetest/v1 JSON envelope schema
• The on-disk Run Record format under .novetest/
• Linux (x86_64, aarch64), macOS (universal2), and Windows (x86_64) distribution

Roadmap:

• Rich TTY renderer — tables, colors, and single-line summaries for human use
• Nove Test Console — a dashboard for humans to inspect what AI coding agents are doing through Nove Test
• Nove Test Team — team-scale collaborative test-driven development built on the Nove Test engine

Documentation

User documentation ships in two parallel sets — same flow, same verbs, tuned to two audiences:

• For humans — install, quick start, per-language notes, deeper verbs, troubleshooting. Examples are scannable terminal text with glyph summaries (✓ ✗ — ⚠ ! ? · ↳).
• For AI agents — same flow, same verbs, but every example is a full novetest/v1 JSON envelope with deterministic routing on recommendations[].category.

Start at docs/ for the audience picker.

License

Nove Test is released under the Apache License 2.0 — see LICENSE.

Free for any use:

• Internal use, CI integration, production deployment, by individuals and organizations of any size
• Forking, modifying, redistributing (with notices preserved)
• Academic and research use, including publishing modifications

The Apache 2.0 patent grant terminates if you initiate patent litigation against Nove Test or its contributors (Apache 2.0 §3).

Enterprise licensing inquiries (custom terms, indemnification, support contracts): admin.nove@gmail.com

Contributing

Pull requests, bug reports, documentation improvements, and adapter contributions for new test engines are all welcome. See CONTRIBUTING.md for how to get started.

---

Stop instant testing.

Build continuous testing intelligence with Nove Test.