Quelltest documentation
Quelltest is a pair of tools that check whether your tests would actually notice a bug. The npm package (quelltest, JavaScript and TypeScript) is a free, offline gate for AI-written tests: it proves a test fails when the code is wrong and fails the build when a change weakens tests. The PyPI package (quelltest, Python) reads docstrings, Pydantic models and type hints, writes a pytest test for each gap and proves each test catches the violation. Pick the package that matches your language below.
Which package should I use?
| npm | PyPI | |
|---|---|---|
| Language | JavaScript and TypeScript | Python 3.11+ |
| Install | npm install -D quelltest | pip install quelltest |
| Command | npx quell | quell |
| Version in these docs | 0.3.1 | 2.0.0 |
| Licence | Apache-2.0 | MIT |
| Test runners | Jest, Vitest, Mocha, node:test | pytest |
| Main job | Mutation-check tests, gate weakened tests, list risky untested edges | Generate and verify pytest tests from specs already in your code |
Both packages are named quelltest and both installed command names are quell. They are separate projects with separate commands, so a repository with Python and Node code can use both.
I write JavaScript or TypeScript
- Install with
npm install -D quelltest. - Follow the npm quickstart: find risky edges, score them against coverage, generate tests, verify a test, gate a change.
- Put the gate in CI with the CI guide, or give it to a coding agent through the MCP server.
npm install --save-dev quelltest
npx quell verify tests/pay.test.js --source src/pay.js --runner jest
npx quell gate --diff origin/main
I write Python
- Install with
pip install quelltest. - Follow the Python quickstart.
- Read how the five gates work, then wire it into CI or give it to an agent through the MCP server and Claude Code plugin.
pip install quelltest
quell find src/
What do the two packages have in common?
- They prove tests, not only coverage. Line coverage says code ran. Both tools check that a test fails when the code is broken: the npm package by mutating your source in a throwaway copy, the Python package by injecting a violation of the requirement it is testing.
- They run on your machine. The npm package never sends code anywhere. The Python rule engine needs no network; the LLM fallback is opt-in. See npm privacy and LLM providers.
- They are built for agents. Each ships an MCP server so a coding agent can run the checks itself: npm and Python.
Where is everything else?
- Python reference pages: the five gates, three-bucket output, Production Readiness Score and constraint kinds.
- Claude connector: read synced readiness reports from Claude.ai (Pro and Team).
- npm limits and known issues: what the npm package does not do.