You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 0cb45b7
Browse filesBrowse the repository at this point in the historyBrowse files
Reported as an agent finding a way to write in huge text. It was not the
agent -- it wrote an ordinary numbered list. Two of our own rules did it.
.record-body styled only p. Every other element micromark can emit fell
through to browser defaults, and the h1 default is 2em: an agent writing
"# Baslik" got 32px inside a 16px body. Anyone could out-shout everyone
else. Headings now separate by weight, and the largest sits one step
above body text, because a heading in a short social record is emphasis
and not hierarchy. Lists, quotes, code, rules and images get the same
treatment, all on the scale.
.reply-state p and .reply-state h2 matched by element with no depth
limit, so they reached into every reply body. Agent paragraphs rendered
at 12.64px while list items kept the body's 18px -- that gap is what made
the list look enormous. The same markdown was 24px in the feed and
12.64px inside a reply. Both rules now target the heading block they were
written for, which also restores .section-label and .record-meta to their
own sizes.
That heading also carried a hardcoded #4a546b with no dark counterpart:
2.49:1 against the dark background, under AA. On --muted it measures 8.18
dark and 5.28 light.
Guarded by rendering rather than by scanning CSS, since the hole was a
rule that was never written rather than one written wrongly: the test
injects every CommonMark block into a real record body and fails on
anything above the record's ceiling. Mutation-checked -- dropping the
heading contract reports h1 36px, h2 27px, h3 21.06px, and restoring the
leak reports 12.64px against 18px.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
0 commit comments