fix(design-sync): match mono/code as whole words in font classifier (CE-7 review)

[linardsb]

Summary

PR #472 was merged before its round-1 review fixes landed. This PR carries them. The main fix is review F1: the font classifier matched mono and code as name prefixes, so sans and display fonts such as Codec Pro, Monotype Corsiva and Monoton got a Courier New fallback. That is worse than the blanket sans-serif used before #472. Now mono/code match only as whole words. The brand stems (courier, consol, menlo, monaco) still match as prefixes, so Consolas stays mono.

What changed

• app/design_sync/font_stacks.py: new _MONO_TOKENS = {"mono", "code", "monospace"} (whole tokens); _MONO_PREFIXES cut down to the brand stems, plus the one-word mono brands monolisa, monoid, mononoki, monofur, codenewroman (fix(design-sync): match mono/code as whole words in font classifier (CE-7 review) #473 review F1). monospace keeps a leading ui-monospace/monospace generic mono (fix(design-sync): match mono/code as whole words in font classifier (CE-7 review) #473 review F2).
• test_font_stacks.py: the reviewer's names Codec Pro, Monotype Corsiva, Monoton, Monofett → sans, Monotype Garamond → serif, Source Code Pro → mono; fix(design-sync): match mono/code as whole words in font classifier (CE-7 review) #473 review: MonoLisa, Monoid, Mononoki, Monofur, CodeNewRoman → mono, and "ui-monospace, Menlo" / "monospace, Foo" pinned to their origin/main stacks. The serif check now uses endswith(", serif"), so sans-serif no longer passes it (F6).
• test_component_renderer.py: test_heading_font_override_quotes_unescaped, which fails if _replace_heading_font goes back to html.escape (F5).
• Renamed the three *escape* font tests to *strip*, since they now check that characters are stripped (F7).
• Plan .agents/plans/ce-7-font-fallback-stacks.md: D3 and the classifier IMPLEMENT line now describe whole-word matching.

Diff: 6 files, 60+/9− (observed, git diff --stat origin/main...HEAD at 94e84b95).

Validation

observed — make check-full via record-gate.sh, base d685e100 plus the uncommitted #473 review fixes (the same app/plan tree committed as 94e84b95; derived from the staged set), exit 0:

pytest:  9137 passed, 0 failed, 123 skipped
vitest:  780 passed, 0 failed
window:  2026-10-04T21:11:36Z -> 2026-10-04T21:17:14Z

• RED: the 5 new F1 cases fail on the unfixed classifier: 5 failed, 24 passed (observed, pytest test_font_stacks.py -k font_category). F5: the reviewer's mutation (html.escape(font_stack(font), quote=True) in _replace_heading_font) → 1 failed (observed).
• test_font_stacks.py has 93 tests (observed, pytest test_font_stacks.py 93 passed). fix(design-sync): match mono/code as whole words in font classifier (CE-7 review) #473 review RED: unfixed classifier with the 7 new cases → 7 failed, 86 passed (observed). 9130 + 7 = 9137 (derived).
• Converter output is byte-identical: the 7 snapshot tests and 7 corpus-category tests pass, and no file under data/ changed (observed). The A3 scores are therefore unchanged (derived from identical output; the scorer was not re-run). None of the corpus first families (Arial, Courier New, Geist Mono, Helvetica, Inter, Noto Sans, Roboto, -apple-system) contains a token that starts with mono/code without being exactly that word.
• Manual check: none needed (classifier and tests only).

Notes for the reviewer

• Scope (user's call): F1 plus the test-only Lows F5, F6 and F7. F2 (missing mono names such as Inconsolata), F3 (dotted names left unquoted) and F4 (combining marks deleted) were dropped as Low: each behaves the same as before fix(design-sync): font fallback stacks by category (CE-7) #472. No ledger change.
• The same commit, 56c31a47, also sits on the merged branch feat/ce-7-font-fallback-stacks. This PR cherry-picks it onto da1524e3, and the tree is identical to 56c31a47.

Linked

Review fixes (round 1 of #473): .claude/reports/pr-473-review-fixes.md (F1, F2).
Review: .claude/code-reviews/pr-472-review.md (F1, F5–F7). Follows #472.

Opened as a draft; CI's ready job flips it when every ci.yml job (backend, frontend, sdk-check, trivy, migrations, integration, migration-lint, commit-lint, e2e-smoke) and codeql are green on this head. A red job leaves it here with the failing check on the PR.

🤖 Generated with Claude Code

[qodo-code-review]

ⓘ Qodo reviews are paused because the subscription is no longer active. Ask your workspace admin to reactivate the subscription to resume reviews. Manage billing

[linardsb]

Verdict: APPROVE (with two follow-ups). Recorded as a comment because GitHub blocks a formal approval of a self-authored PR.

PR #473 review — match mono/code as whole words in font classifier (CE-7 review)

Head d685e100 · Base main @ da1524e381d1ec7957328eaf76ec49a5d4bebd83 (live origin/main tip at review time = same sha) · round 1

Summary

Carries the PR #472 round-1 fixes F1, F5, F6, F7. The F1 fix is correct for the names it targets: Codec Pro, Monotype Corsiva, Monoton, Monofett now get a sans stack and Monotype Garamond a serif stack. The change has one side effect the PR body does not mention: real monospace fonts whose name is a single word starting with mono, plus a CSS generic monospace in first position, now also fall to sans. Neither is in the corpus (converter output is byte-identical, observed), so neither blocks merge.

Issues

Critical — none. High — none.

Medium

F1 · app/design_sync/font_stacks.py:57,102 — single-word monospace fonts now get a sans fallback (undocumented).
Whole-word mono drops every mono font whose name is one token beginning with mono (observed, old vs new font_category run against origin/main:font_stacks.py):

Family	origin/main	this PR
MonoLisa	monospace	sans-serif
Monoid	monospace	sans-serif
Mononoki	monospace	sans-serif
Monofur	monospace	sans-serif
CodeNewRoman	monospace	sans-serif

The PR body says only that Consolas stays mono; it does not state that these move. The behaviour equals pre-#472 (everything was sans), which is why this is Medium, not High.
Fix: add the stems to _MONO_PREFIXES (monolisa, mononoki, monoid, monofur) with parametrize cases, or record the trade in the PR body and plan D3. Prefix monoid does not catch Monotype/Monoton/Monofett.

Low

F2 · app/design_sync/font_stacks.py:102,131 — a generic monospace in first position now classifies as sans.
font_stack short-circuits only when the last family is generic. A list that starts with one and does not end with one reaches font_category, which no longer matches monospace (observed):

• "ui-monospace, Menlo" → old …, 'Courier New', Courier, monospace; new …, Helvetica, Arial, sans-serif.
• "monospace, Foo" → same flip.

Low because the input needs a hand-authored list that starts with a generic and lacks a trailing one; no current producer emits that.
Fix: add monospace to _MONO_TOKENS (matches ui-monospace via the hyphen split) plus a test case.

Reviewer agent also proposed Nanum Gothic Coding as a regression; rejected: "coding".startswith("code") is false, so it was sans on origin/main too (observed).

Numbers pass

Claim (PR body)	Check	Result
6 files, 29+/9−	gh pr view additions/deletions, git diff --stat origin/main...HEAD	matches (observed)
make check-full exit 0: 9130 passed, 123 skipped; vitest 780	re-ran make check-full on d685e100	9130 passed, 123 skipped, 7 xfailed; vitest 780; exit 0 (observed)
test_font_stacks.py has 86 tests	pytest --collect-only	86 (observed)
RED: 5 failed, 24 passed with -k font_category on unfixed code	-k font_category collects 29; old classifier returns MONO for exactly the 5 new F1 names, SANS-expecting none else	consistent: 29 − 5 = 24 (derived, not re-run)
F5 mutation → 1 failed	reviewer agent read the test: html.escape produces ', failing both asserts	consistent (derived, mutation not re-run)
Tree identical to 56c31a47	git diff --stat 56c31a47 HEAD	empty (observed)
A3 unchanged	derived from byte-identical output; correctly labelled derived	OK

Validation

Gate	Result
make check-full @ d685e100	exit 0: ruff clean, mypy "no issues in 1398 files", pyright 0 errors, pytest 9130 passed / 123 skipped, vitest 780 passed, flag audit 0 errors / 14 warnings, .env.example in sync (observed)
CI (gh pr checks 473)	all 14 checks pass, incl. backend, frontend, CodeQL, Semgrep (observed)
Code-scanning alerts on refs/pull/473/merge	0 open (observed)
Deferred items	no entry references font_stacks, CE-7 or font_category

Side note: make check-full rewrites the date in app/ai/agents/{dark_mode,scaffolder}/skill-versions.yaml to today. Not this PR's concern; restored after the run.

Done well

• Three-line production change; the comments on _MONO_TOKENS and _MONO_PREFIXES say why each is whole-word or prefix.
• Monotype Garamond → SERIF also proves the serif branch still runs once mono is ruled out.
• test_heading_font_override_quotes_unescaped pins the exact output string and the absence of ', so a revert to html.escape fails it.
• F6's endswith(", serif") closes the sans-serif false pass.
• Plan D3 and T2 updated to match the code.

Recommendation

Approve. F1 is a two-line fix plus four test cases and is worth doing in this PR or a follow-up; F2 can ride with it.