Test Coverage
Atlas measures test coverage across every package in one run. This page covers the command and how to read the results.
pnpm run build
pnpm run test:coverageMerged coverage per package (every package test run counted)
Package Lines Branches Functions Statements
---------- ------ -------- --------- ----------
core 92.6% 82.5% 92.6% 91.7%
drive 93.3% 88.1% 93.9% 92.6%
...test:coverage runs every package's Vitest config as a project of one root run (vitest.config.ts at the repository root), measures every file under packages/*/src, and prints one row per package. The HTML report is in coverage/index.html and the per-file JSON in coverage/coverage-summary.json. CI runs the same command and uploads coverage/ as the coverage-report artifact.
Which number to read
Read the merged row above. Shared code is mostly exercised through other packages: drive through onedrive and sharepoint, core and types through every package. A package's own run (pnpm --filter @wisecom/atlas-drive exec vitest run --coverage) counts only that package's tests. That run reported drive at 26% and types at 16%, both above 90% in the merged run, and listed files at 0% that other packages test thoroughly. Those numbers point work at the wrong files.
Use a per-package run only to see what one package's tests reach on their own, for example when deciding whether a helper deserves tests next to it.
Reading a gap
Low numbers are a lead, not a verdict. Before writing tests for a file, check the per-file entry in the HTML report and the lines it marks uncovered:
- An adapter that talks to Graph or S3 is exercised only by its own tests, because services mock it. A gap there means the code that runs against the real service is untested. Issue #444 was this case for the drive restore adapters and
fetch_delta. - A file at 0% that nothing in
srcimports is dead code, not a missing test. Remove it with its barrel export rather than testing it. index.tsbarrels andpackages/types/src/testing/are excluded from the measurement.

