Repository navigation
Description
Finding
docusaurus.config.ts's markdown.preprocessor strips .md from
site-internal links using a regex with two syntax alternatives:
/(\]\(|\]:\s*)(\/(?:docs|analyses)\/[^\s)#?]+)\.md(?=$|[\s)#?])/g](— inline links, e.g.[docs](/docs/foo.md)]:— reference-style link definitions, e.g.[1]: /docs/foo.md
lib/docusaurusConfigPreprocessor.test.mts only exercises the ](
branch. c8 reports 100% statement/branch coverage for this file because
both alternatives live inside one regex literal, which the coverage
instrumenter does not decompose — so the gap is invisible to the
existing npm run test:coverage gate.
Evidence (manual mutation, reproducible): removing the |\]:\s*
alternative from the regex and re-running
node --experimental-strip-types --experimental-test-module-mocks --test lib/docusaurusConfigPreprocessor.test.mts
against revision 418adeb still passes all tests. No end-to-end suite
exists in this repository to catch it either (there is no e2e/
integration harness here — see techdocs#390 for the broader CI-signal
gap), so this is Priority 1 (covered by neither unit nor end-to-end
tests) under the quality lane's coverage-evidence rules.
Recommendation
Add a unit test case in lib/docusaurusConfigPreprocessor.test.mts
that feeds the preprocessor a reference-style link definition (e.g.
[1]: /docs/foo.md) and asserts the .md suffix is stripped the same
way inline links are. A PR implementing this is attached.
Priority
- Impact: medium — a regression in the reference-link branch would
silently leave broken.md-suffixed links on the live site (README
uses this exact import path) with no test or coverage signal to catch
it. - Effort: low — single new test case, no production code change.
Filed by quality agent (hold-gated mode)
🐝 Hive Agent: quality | Instance: hosted-available-lke648397-260827-boq9 | SHA: e9eb8c6
— hive: agent=quality backend=copilot model=claude-opus-4-6 copilot=1.0.88
Activity