kensho-pytest

A pytest plugin that emits the canonical Kensho v1 JSON format. Run your tests, then point the kensho CLI at kensho-results/ to get a self-contained static HTML report.

Install

pip install kensho-pytest
# or, in this monorepo:
pip install -e packages/pytest

The plugin auto-registers via the pytest11 entry point — no conftest.py edits required.

Run

pytest
# => kensho-results/run.json + kensho-results/cases/*.json

# Generate the HTML report (uses the JS CLI from the same monorepo):
npx kensho generate
npx kensho open

CLI flags

--kensho-output PATH                Output dir (default: ./kensho-results)
--kensho-project-name STR           Project name in run.json
--kensho-project-slug STR           Project slug (lowercase, alnum + dash/underscore)
--kensho-run-id STR                 Override the auto-generated run id
--kensho-no-severity-from-marks     Don't promote @pytest.mark.<severity> to case.severity

pytest.ini / pyproject.toml [tool.pytest.ini_options] keys are also respected: kensho_output, kensho_project_name, kensho_project_slug.

What it produces

• kensho-results/run.json — run manifest (project, env, totals, timing).
• kensho-results/cases/<stableId>.json — one file per test case.
• kensho-results/attachments/<caseId>/... — files registered via kensho_pytest.attach.

Each case gets a stable id (tc_<16 hex>) hashed from its nodeid + file path, so test history correlates across runs and across adapters (the JS adapters use the same FNV-1a-based hash).

Markers we read

Inline @tag syntax in test names (def test_foo_at_smoke() or pytest.param(..., id="x@smoke")) is also picked up as a tag, mirroring the JS adapters.

Helper API

Import the helpers from kensho_pytest. All four are no-ops outside a running test, so it's safe to call them from shared utility code.

import kensho_pytest as kensho

def test_login(page):
    with kensho.step("open the login page"):
        page.goto("/login")

    with kensho.step("submit credentials"):
        page.fill("#user", "demo")
        page.click("text=Sign in")
        # nesting works — child steps roll up under the parent
        with kensho.step("verify redirect"):
            assert page.url.endswith("/home")

    kensho.label("team", "growth")
    kensho.link("https://jira.example.com/browse/PROJ-123",
                kind="jira", label_text="PROJ-123")

    page.screenshot(path="/tmp/login.png")
    kensho.attach("/tmp/login.png", kind="screenshot")

Captured output → logs

Pytest's captured stdout / stderr / caplog for each phase (setup, call, teardown) is converted to entries in case.logs[]. The t field is the millisecond offset from the test start.

Errors

call failures map to status: 'fail'; setup / teardown failures map to status: 'broken' (these are infrastructure, not real assertion failures). Skips at any phase map to status: 'skip'. The full longrepr is preserved in case.errors[].stack; the first line becomes case.errors[].message.

Environment auto-detected

GitHub Actions, CircleCI, GitLab CI, Jenkins, Buildkite, Azure DevOps — CI provider, branch, commit, run URL, OS, architecture, Python version.

Pass KR_AUTHOR, KR_COMMIT_MSG, KR_STAGE, KR_BASE_URL, KR_APP_VERSION, KR_BUILD_NUMBER, KR_RELEASE, KR_REGION, KR_LOCALE, KR_TRIGGER, or KR_FEATURE as env vars to populate the matching fields on run.env.

Design notes

• Zero runtime dependencies beyond pytest itself. The schema lives in the JS workspace; we vendor the minimum (id-hashing, status mapping, env capture) inline so the adapter installs in seconds.
• The reporter never raises out of a hook — a broken adapter must not break a test run. All errors come out as warnings.warn.
• IDs are stable across adapters: the FNV-1a hash matches the JS stableCaseId byte-for-byte, so pytest and playwright runs of the same suite roll up to the same history on the platform.

License

Apache-2.0.