v0.49.3

v0.49.3

• Updated the wordpress.org listing screenshots (.wordpress-org/screenshot-1.png
  through -5.png) to the current settings panel — the previous set was several
  versions out of date (v0.35.2) and missing panel sections shipped since
  (Extra custom fields, Excluded builder elements, the hit counter's named-bot
  breakdown). Added a sixth screenshot showing the [sysmda_md_actions]
  front-end split button (copy/view/download), with a matching caption in
  readme.txt. Metadata and repository tooling only — no plugin code, output
  or behaviour changes.

v0.49.2

v0.49.2

• The Live Preview blueprint (.wordpress-org/blueprints/blueprint.json) now
  lands on the site's front page (/) instead of /hello-world.md. Landing
  directly on a text/markdown response gave a first-time visitor a download
  or a wall of unstyled plain text with no context — not a demo. The seed
  content every fresh WordPress install carries (the "Hello world!" post, the
  "Sample Page" page) already exercises both enabled post types and
  /llms.txt, and a real front page a visitor can click around makes a
  better first impression. sysmda_llms_txt_enabled is now also set
  explicitly in the blueprint (already on by default, but demonstrated
  regardless of that default). Metadata and repository tooling only — no
  plugin code, output or behaviour changes.

v0.49.1

v0.49.1

• The plugin now ships a .wordpress-org/blueprints/blueprint.json, enabling
  wordpress.org's Live Preview button on the plugin's listing page:
  visitors can try the plugin live, active, in WordPress Playground (a
  WASM WordPress running in the browser) with no install. The blueprint
  installs the plugin from its own published wordpress.org release, enables
  the post and page content types, sets pretty permalinks and lands on
  the seed post's .md — the most direct demonstration of what the plugin
  does. Metadata and repository tooling only — no plugin code, output or
  behaviour changes; not bundled into the plugin zip.

v0.49.0

v0.49.0

• Added: pages now point at the site's /llms.txt with the rel="describedby" link relation introduced by version 2 of the llms.txt specification, in both the HTML head and the HTTP Link: header — alongside the Markdown alternate they already advertised. An agent landing on any article can find the site's index without having to guess that /llms.txt exists. Emitted only where the Markdown alternate is already emitted and only while this plugin's own /llms.txt is enabled, so the link never points at an endpoint that is not being served; there is no new setting.
• Changed: MarkdownController::link_header_has_alternate() is now link_header_has_relation() and takes the relation to look for. It is an internal helper, public only so its parsing can be exercised from the test suite; no documented filter or output contract refers to it.

v0.48.0

v0.48.0

• Added: a per-known-bot-name breakdown in the .md hit counter. Below the
  existing bot/human totals, a second table names any of a short curated list —
  ClaudeBot, GPTBot, PerplexityBot, CCBot, matched together with their
  user-initiated variants (Claude-User, ChatGPT-User, OAI-SearchBot,
  Perplexity-User) — with at least one hit in the last 30 days. A crawler
  never seen gets no row at all, rather than a permanent zero. Still
  aggregate-only and count-only: it names a few crawlers already counted
  inside the bot total, it does not add any new stored data, IP address,
  raw user-agent string or per-visitor identifier. New filter
  sysmda_md_hits_named_bot_patterns (Advanced), independent of the existing
  sysmda_md_hits_bot_patterns — one decides bot vs human, the other only
  names a match already counted as a bot.

v0.47.1

v0.47.1

• Fixed: a text custom field containing Markdown punctuation was published with
  that punctuation active — a field reading A *literal* marker arrived with one
  word in italics instead of the asterisks the author typed. Underscores,
  brackets and backslashes were the same case. Each value was wrapped in a
  <div>, and that wrapper silently switched off the escaping every other piece
  of text in the document gets. Values are now separated by a blank line instead,
  which restores the escaping and keeps them apart; markup from a WYSIWYG field
  is unaffected and still converts as before.
• Fixed: listing a custom field whose value contains Gutenberg block markup
  alongside plain-text fields ran those text fields together on one line. One
  block-valued field sent every sibling down the block path, where plain text is
  emitted without paragraphs.

v0.47.0

v0.47.0

• Added: Extra custom fields — a new setting listing the post meta keys whose
  values belong in the Markdown. Their content is appended to the end of the
  body, in the order listed. One setting covers ACF, JetEngine, Meta Box and
  WordPress's own Custom Fields box, because underneath they all store post meta,
  so a page whose text comes partly from a template's fields is no longer
  published half missing. Empty by default and never detected automatically: a
  field starts appearing when you type its key into the box, and not before.
  Values that are not text — an image, a repeater, anything stored as an array —
  are skipped rather than guessed at.
• Added: sysmda_markdown_extra_meta_keys (Stable) as the filter behind the new
  setting, so the list can be varied per post from code.
• Changed: a post that does not carry any of the configured keys keeps its
  document and its cache validator byte-identical, so configuring a field for a
  couple of landing pages does not make every article on the site revalidate.
• Fixed: ACF field values added through sysmda_acf_field_keys never reached a
  Bricks page's Markdown. Such a page is rendered through Bricks' own API, which
  does not read the post content the values were appended to, so they were
  dropped silently — since 0.46.0. Both they and the new custom fields now go
  through a dedicated sysmda_markdown_appended_html filter (Advanced) that is
  honoured on every render path.
• Fixed: the Excluded builder elements setting added in 0.46.0 was not removed
  on uninstall.

v0.46.1

v0.46.1

• Fixed: an ACF subtitle containing Markdown punctuation was parsed instead of
  read. A subtitle of A *literal* marker was emitted raw between the emphasis
  delimiters, so the reader's own asterisks took part in the formatting and the
  single italic line came out as three runs of emphasis. Subtitle text is now
  escaped the same way the article body is, so it reads back exactly as typed —
  asterisks, underscores, brackets, backslashes and a leading # included.
• Fixed: a link whose body is a non-breaking space (or another invisible
  character) is now named from its aria-label or title like any other link
  that renders nothing. Card and link-preview plugins fill an overlay link with
    rather than leaving it truly empty, and those links were still
  arriving as [](url "Name") — with the name in the tooltip and nothing tying
  it to the address.
• Changed: the documentation site is now built on every pull request instead of
  only after merging, so a broken page cannot reach the published site.

v0.46.0

v0.46.0

• Bricks pages now get a real .md (Phase 2 of the page-builder plan;
  bricks leaves BuilderDetector::AWAITING_ADAPTER, which is the whole
  phasing mechanism — no other rule changed). The new BricksAdapter renders
  a Bricks-mode post through Bricks' own \Bricks\Frontend::render_data()
  rather than reimplementing its element types, the same "read the builder's
  data to decide, read the vendor to render" shape BuilderDetector already
  used. A new BuilderAdapter interface and a third branch in
  ContentRenderer::render(), tried before the block/classic branches, is the
  seam: sysmda_markdown_builder_adapters (Advanced) is the extension point.
  A post switched back to Render with WordPress is unaffected — the adapter
  requires the render mode AND the Bricks plugin active, so with either
  missing the ordinary post_content pipeline is the correct answer, exactly
  as it was before this release.
• Fixed the one defect the Phase 0 reconnaissance found: Bricks' own image
  lazy-loading swaps src for an inline SVG placeholder and moves the real
  URL to data-src unless \Bricks\Database::$page_settings['disableLazyLoad']
  is set for the render — the adapter now brackets every render with a
  save/restore of that flag, so every Bricks image converts to a normal
  ![](url) instead of a meaningless data URI.
• New exclusion list: sysmda_markdown_excluded_builder_elements
  (Stable), a panel field and filter for page-builder chrome — Bricks'
  form, nav menu, share bar, table of contents and breadcrumbs elements are
  excluded by default, the same way sysmda_markdown_excluded_classes
  excludes a CSS class. Additive to the built-in defaults, never a
  replacement (the 0.40.0 rule). The existing md-exclude class already
  worked on Bricks elements with no code change needed — Bricks emits it
  verbatim on the element's wrapper.
• The cache validator now covers Bricks content: the render mode, a hash
  of the stored element tree, and the modification date of any referenced
  template element's own post are folded into the existing dependency
  fingerprint, so a mode flip or a template edit moves the ETag even though
  post_modified_gmt does not.
• The front-matter description fallback and /llms.txt entries now have
  a Bricks-aware last resort, after Rank Math and the excerpt: a cheap,
  unrendered read of the stored tree's text-bearing settings, run through the
  same exclusion pass as the body. A Bricks post's post_content is never
  used for this, even when the adapter's fallback finds nothing — it can hold
  stale prose left over from before the page was rebuilt in Bricks, and
  publishing that would reproduce, in the description field, the exact
  "confidently wrong" failure the page-builder veto exists to prevent in the
  body.
• Advanced-level toggle: sysmda_markdown_builder_suppress_content_filters
  (default on) removes foreign the_content callbacks — a related-posts
  block, a CTA — while Bricks' own post-content element renders, since that
  element calls the full the_content chain internally and would otherwise
  reintroduce exactly the injected content this pipeline avoids everywhere
  else. Flagged as a maintainer-reversible design choice; return false from
  the filter to accept whatever a real visitor sees there instead.
• sysmda_markdown_prewarm stays off for Bricks posts as for everything else:
  element-level visibility conditions are unverified under WP-Cron's missing
  request context (documentation only, no code change).
• Elementor remains the only builder in AWAITING_ADAPTER; Divi, WPBakery,
  Oxygen, Beaver Builder and Breakdance stay permanently unsupported.

v0.45.1

v0.45.1

• Fixed: the [sysmda_md_actions] dropdown opened off to the right of the
  button instead of below it. The menu was anchored to the caret rather than to
  the split button as a whole, so its left edge started halfway across the
  control and the panel hung out over the text beside it. It is now aligned to
  the button's own edge and drops straight below, which is what the control
  looked like it should do all along.
• Fixed: the fallback placements now run only when they are needed, and they
  handle the awkward cases better. The menu switches to the opposite alignment
  when the button sits too close to the screen edge, flips above the button when
  there is no room below, and — where there is room on neither side — caps its
  height and scrolls instead of growing across the button it belongs to. On a
  right-to-left site both alignments mirror.
• Changed: the menu is sized to its own content instead of a fixed 250 px, with
  the same width as its floor, so a longer translated label is no longer squeezed
  into two lines while a narrow screen still keeps it inside the viewport.