0.10.1

Bug Fixes

• design-validate now reports the true number of hard-wrapped lines in a document. The detector capped its counter at five warnings per file and stopped counting there, so a document hard-wrapped end to end was indistinguishable from one with five bad lines — a 533-line doc that needed reflowing throughout reported the same "5" as a doc needing five edits, and readers scoped their fix to the five lines shown. Line-level detail is still capped at five to keep the report readable, but the count is now uncapped and a trailing N hard-wrapped lines total, first 5 shown summary fires whenever the cap is exceeded. The aggregate **Warnings:** tally reflects the true total as well, since the caller folds the same counter into it.

  This also resolves the companion report that wrapped list-item continuations were never flagged. They were in fact detected all along — the detector's block-marker heuristic treats an indented continuation under a list item exactly like a wrapped paragraph, while correctly skipping nested list items, fenced code, tables, and block quotes. What actually happened is that paragraph-level wraps earlier in the file exhausted the five-warning budget before any list-item wrap could print, so the list warnings were starved rather than missing. Making the total visible surfaces them. #76

Performance

• design-link classifies references in a single awk pass instead of forking per reference. The checker was subprocess-bound: it forked a grep per reference to test node-set membership, a subshell per path resolution, and re-parsed a target document's heading slugs once per anchor link pointing at it — upwards of 1,600 process spawns on a 22-document corpus, scaling with reference count rather than corpus size. A link checker only catches link rot if people are willing to run it, and at roughly 45 seconds it had stopped being something anyone ran casually.

  Collection now extracts raw reference records only, and one awk invocation classifies the whole corpus — path resolution, node membership, and the heading-anchor pipeline included, with heading slugs memoized per target file. Associative arrays are available inside the awk program regardless of the bash 3.2 constraint that governs shipped scripts. A synthetic 22-document corpus went from 29.6s to 2.0s, with byte-identical output. Detection semantics and report format are unchanged.

  The node list is passed to awk as a file argument rather than through -v, because the BWK awk that ships as /usr/bin/awk on macOS rejects a -v value containing a newline — which would have silently classified zero references on a stock Mac. #76

Patch Changes

Thanks to @spencerbeggs for their contributions!

This is a version-only release. No packages were published to a registry.