pytest-xspec

Executable specifications with pytest.

Link specifications to your existing pytest tests, organize your test suite your way, and see which specifications your test suite verifies. Detect missing tests and enforce verification in CI.

Why pytest-xspec?

As software engineers, we like code. Ask us what the software does, and we might say, "Just read the code." But stakeholders need a shared understanding of what the software should do. We turn to acceptance tests to make those expectations precise, using examples in the language of the business.

For many of us, making a whole test suite read as a clear specification is costly. It takes time and practice, and the work continues as the codebase grows. Related behavior spreads across files, implementation checks sit beside acceptance tests, and the overall requirements become harder to see.

pytest-xspec connects a readable overview of your requirements to the tests that verify them. Keep specifications in a document your team can discuss, put tests where they make sense, and link them with IDs. The report shows which requirements are verified, which remain unverified and which have no tests. When business asks what is covered, you have an answer they can read.

Usage

Install pytest-xspec into your pytest environment. The plugin is discovered automatically. Python 3.10+ and pytest 8+ are supported.

pip install pytest-xspec

Write your specifications

Create xspec.toml in pytest's project root, usually the top-level project directory. Keep each specification, its description and the reason it matters together in one document.

[[requirements]]
id = "REQ-001"
description = "Adding two integers returns their sum."
rationale = "Calculations must give predictable results."

[[requirements]]
id = "REQ-002"
description = "Adding zero leaves the value unchanged."

Use one [[requirements]] entry per specification. Each entry needs an id. The description states the requirement, and the rationale explains why it matters. Both are optional and can be empty strings. This text stays in the document, while tests and reports refer only to the IDs.

See the specification document example for a complete runnable project.

For specifications maintained in other documents or tables, declare just their IDs.

Mark tests

Associate tests with specifications, using one marker per specification.

import pytest


@pytest.mark.xspec("REQ-001")
@pytest.mark.xspec("REQ-002")
def test_adding_zero_preserves_the_value() -> None:
    assert 2 + 0 == 2

Run pytest

Plain pytest checks specification configuration and markers without printing a specification report. Add --xspec-report to see the report, or --xspec-strict to also fail the run when a specification is unverified.

pytest --xspec-report             # Print the verification report
pytest --xspec-json=report.json   # Export statuses and individual test outcomes
pytest --xspec-strict             # Report and fail on unverified specifications
pytest -m xspec --xspec-strict    # Run only marked specification tests

The report is green when all specifications are verified and red when verification fails. The first report shows the example above; the second shows a specification without tests.

Select specifications

Use --xspec-select=ID to run tests linked to one specification. Repeat the option to select several specifications. A test linked to several selected IDs runs once.

pytest --xspec-select=REQ-001
pytest --xspec-select=REQ-001 --xspec-select=REQ-002

Each option takes one exact ID, including any spaces or commas. Quote IDs when the shell requires it. Unmarked tests are excluded. Collection paths, -k and -m further narrow the selection. An undeclared selected ID is a usage error. Selecting only IDs without matching tests produces an empty selection.

Strict mode still evaluates every declared specification. Selecting one ID keeps other specifications in the report and the coverage calculation. Excluded tests remain DESELECTED, and tests shared with other specifications keep all their links. With selection active, terminal reports and the JSON scope field show selected IDs, collection paths, -k, -m and whether the run only discovers tests.

Discover tests without running them

Add --collect-only -q to list the selected test IDs, including parameterized cases.

pytest --xspec-select=REQ-001 --collect-only -q

Discovery runs neither fixtures nor test bodies. It still imports collected modules, which need their normal dependencies. Reports mark discovered tests as SKIPPED, so discovery does not verify their specifications. Strict mode fails when any specification remains unverified. Without strict mode, an empty selection keeps pytest's no-tests exit code, 5.

Read specification statuses

Reports label specifications as requirements.

A specification is verified when every test marked with its ID passes. One passing test is not enough if another test for the same specification fails or skips.

When those tests have different results, the report uses the first matching status in this table, reading from top to bottom.

For example, if one test passes and another skips, the specification is SKIPPED and remains unverified. If one test fails and another skips, it is FAILED.

Missing and unfinished tests

Pytest first finds tests, then runs them. If it finds no marked tests for a declared specification, that specification is MISSING. Tests that were found but never ran are SKIPPED, including runs that only find tests or stop early. If pytest encounters an error while finding tests, such as a broken import, the verification check cannot succeed.

Tests without markers

Tests without a specification marker do not change specification statuses. They can still make pytest fail. In the duck example, the specification passes, but the unmarked cheese test fails the pytest run.

Check coverage

Coverage is the percentage of declared specifications that are verified. With no declared specifications and no errors while finding tests, the verification check succeeds and coverage is 0.0%.

Errors while finding tests make verification fail, even at 100% coverage. The report follows pytest's color settings, so --color=no disables color.

For a complete report, let pytest find the whole test suite. With -k or -m, the plugin still knows about the tests excluded by the filter and reports marked ones as DESELECTED. If you run only one folder, it cannot see tests in other folders.

Use an existing specification

If you maintain specifications elsewhere, declare their IDs in pyproject.toml or pytest.ini. Your original document remains the source for descriptions and context. The markers, commands and reports work the same way with either configuration format.

In pyproject.toml, use a list of IDs:

[tool.xspec]
requirements = ["REQ-001", "REQ-002"]

In pytest.ini, use one ID per line:

[pytest]
xspec =
    REQ-001
    REQ-002

How the source is chosen

An ID list in either configuration file takes precedence over xspec.toml. Declaring IDs in both pyproject.toml and pytest.ini is an error. Specifications from different sources are never merged.

When neither configuration file declares IDs, pytest-xspec reads xspec.toml from the project root. Those configuration files can still contain other settings. An explicitly empty ID list also takes precedence, so it declares zero specifications even when xspec.toml contains entries. An unused xspec.toml is not read or validated.

Empty specifications and validation

IDs are strings and are used exactly as written. Duplicate IDs and markers referring to undeclared IDs are rejected. In xspec.toml, supplied descriptions and rationales must also be strings. Additional fields are ignored.

To declare zero specifications in xspec.toml, write requirements = []. A document without requirements, including an empty file, is a configuration error. Invalid documents produce an error before tests run. If no source exists, there are zero declared specifications.

Runnable examples

Start with the pizza, then try the other examples to see what makes a specification pass or remain unverified.

Run the commands below from the repository root. Dependencies are installed automatically.

Run an example

Run the Quickstart example, then replace quickstart with another example folder.

uv run --locked pytest examples/quickstart --xspec-strict

For specification_document, the requirements live in xspec.toml. Its pyproject.toml only sets test discovery, leaving the document as the requirement source.

For acceptance_only, add -m xspec to run only the marked test. Without that filter, the unmarked test makes pytest fail.

Select and discover example tests

Use the same example command with --xspec-select=REQ-001 to select the pizza test. Add --collect-only -q to discover it without running it:

uv run --locked pytest examples/quickstart --xspec-select=REQ-001 --collect-only -q

For missing_requirement, repeat --xspec-select=REQ-002 to select both declared IDs. Only the existing sock test runs. With --xspec-strict, the missing sock still makes verification fail.

Inspect the JSON report

Export the example results to see the specification status and individual test outcomes.

uv run --locked pytest examples/quickstart --xspec-json=examples/quickstart/report.json

Open examples/quickstart/report.json after running the command.

Verify all examples

Check that every example behaves as described.

uv run --locked pytest examples/test_examples.py -n 0

All six checks should pass, including checks for the examples that deliberately fail. The examples are also verified in CI.

Development disclosure

This package is developed by AI agents under human direction. See AI_DISCLOSURE.md for details on authorship, human review and repository workflows.

This README was written by AI.

Contributing

See CONTRIBUTING.md for setup, tests and pull requests.