v0.11.0

Added

• GFM alerts ↔ org special blocks: > [!NOTE] ↔ #+begin_note, a
  title on the marker line ↔ the block's parameters
  (> [!TIP] Stretch first ↔ #+begin_tip Stretch first). A query
  block, and a type Obsidian's callouts read otherwise (a.b), stay
  org text; Logseq md keeps writing special blocks as org does.
• Obsidian md writes an alert as a callout, its type lower case
  (> [!note]), and keeps a callout's fold as the block's parameters'
  leading token (> [!tip]- T ↔ #+begin_tip - T).
• The docs are on the site: docs/ and CONTEXT.md render as pages under
  /docs/, listed on a generated index linked from every page's nav
  (ADR 0009); a link between docs stays on the site, and code is
  highlighted in binarypoetry.ch's colors. The text keeps VitePress's
  reading width, about 90 characters a line, and on a wide screen a
  page's sections are listed beside it, the one being read marked,
  and on a wider one every doc.
• The Web UI lists the licenses of the third-party packages it is
  built from, runtime and development, on a licenses page
  (licenses.html), with each package's license text (ADR 0008),
  linked from the footer of the landing and converter pages.
• An image's size has a Spelling, images (ADR 0007): "html"
  writes an image link sized by #+ATTR_HTML: :width/:height as
  <img src alt width height> and reads that <img> back; under
  "markdown", the default, the attribute line stays verbatim.
• Obsidian md reads an image's size (![alt|300](image.png),
  |300x200) as its #+ATTR_HTML: :width/:height line and writes
  it back, for an image alone in its paragraph outside a list;
  translating Obsidian md to or from Vanilla md turns one into the
  other. An embed's size (![[image.png|300]]) stays as written.
• Logseq's image size ([[image.png]]{:width 300}, in Logseq md
  ![alt](image.png){:width 300}, :height too) is the image's
  #+ATTR_HTML: line in Vanilla org and back, for an image line below
  a block's title that is a paragraph of its own. A block's title is
  a headline, which org cannot size, so a title image keeps its size.

Changed

• Breaking: options are named by format and side (ADR 0007). The
  config's [markdownToOrg] and [orgToMarkdown] sections are gone,
  with an error naming their replacements: [markdown],
  [markdown.input], [markdown.output] and [org.output].
  markdownStyle is [markdown.output.style], recordStyle is
  [org.output] recordMarkdownStyle (CLI --record-markdown-style),
  and the library options are renamed to match (style,
  recordMarkdownStyle).
• Breaking: useHtml and interpretHtml give way to a Spelling
  per construct (definitionList, underline, superscript,
  subscript): "markdown" or "html", set for both sides in
  [markdown] or per side. The library's spelling and per-construct
  interpretHtml replace them; CLI --html replaces
  --interpret-html and sets both sides.
• Breaking: org descriptive lists are written as Markdown
  definition lists (term / : definition) rather than as their
  org text (- term :: definition), and read back from them. A
  Markdown list item holding :: stays a list item, and a
  fixed-width line's : is escaped (\: text), as a line below text
  that starts with a colon now starts a definition. A descriptive list
  nested in a list item does not survive a round trip as one (a known
  limitation of the Markdown parser): it comes back as the item's text.
• Translating Markdown honours the Markdown options, as normalizing
  does: a block holding a construct they concern is converted on its
  own, in the Spelling they name; the rest stays as written.
• The Web UI's options are grouped by the side they act on, showing
  only those of the sides the conversion has (normalizing Org goes
  through Markdown, so its HTML options show too), and a compact table
  sets each construct's HTML reading and Spelling.
• The Web UI's demos show a special block as a GFM alert and a link to
  a headline, and Obsidian Markdown has its own demo: wikilinks, a
  callout, an image's size and a task.
• The Web UI's landing page keeps its text to the docs' reading width,
  about 90 characters a line, centered as the docs' text is, and lists
  what morg does in four points instead of one long paragraph; the
  detail is in the docs.
• The Web UI and the docs draw arrows (→, ↔) outside code in
  DejaVu Sans, subset to them, centered on the lowercase letters:
  Arial's sat low and small beside them. DejaVu's license is on the
  licenses page.
• Logseq org and Logseq md write a block's begin and end lines in
  upper case (#+BEGIN_SRC sh, #+END_QUOTE), as Logseq does; org →
  md → org changed them to Emacs's lower case (#+begin_src), a
  special block's to mixed case (#+begin_TIP). Vanilla org keeps
  lower case.

Fixed

• Obsidian md → org reads a %%comment%% as an org comment (# …)
  alone in its paragraph, else as an html snippet
  (@@html:<!--comment-->@@), and an inline footnote ^[note] as a
  footnote, numbered on from the page's own; both were plain text,
  and the footnote came back escaped (^\[note]), no footnote in
  Obsidian. Org → md writes an inline comment as Obsidian's own
  (%%comment%%), others as an HTML comment, and a footnote. Md → org
  and Obsidian md → Vanilla md read no comment in a wikilink or a diary
  timestamp's <%%(.
• Two quotes next to each other in a list item or a definition stay
  two: Markdown read them, written without a blank line between them,
  as one.
• A special block's parameters (#+begin_tip Stretch first) survive
  org → md: uniorg reads and writes none, so they were dropped without
  a warning; morg reads them from the block's begin line.
• An org descriptive list no definition list can hold (an item with a
  checkbox, or without a term) survives a round trip again: its
  - term :: definition text is written below a
  <!-- morg_descriptive_list --> comment, which reads it back as
  descriptive; it came back a plain list.
• Spelled in HTML, such a list is written the same way: as a <dl>, it
  lost its checkboxes and gained empty terms. So is one whose
  definition holds blocks (an image, a nested list, a second
  paragraph), which a <dl> dropped.
• Reading <sup> or <sub> as org (interpretHtml) keeps the HTML
  where a blank precedes it (x <sup>s</sup>): org reads a script
  only after a non-blank character, so it came back as the text
  ^{s}.
• List item text after a quote or table (- a\n > b\n\n c) stays
  apart from it: org → md wrote no blank line between them, so
  Markdown read the text into the quote, gaining a > each round
  trip, or as a table row.
• HTML in a list item (org's #+begin_export html) stays an HTML
  block in Markdown, and so does text after it: org → md kept the
  item's indentation in the HTML, which Markdown then read as text.
• HTML in a Markdown list item keeps its own indentation in org: md →
  org lost up to the item's indentation from each line, each round
  trip.
• Markup right after other markup (`a`*,*) comes back from org
  without a stray &#xNAN; between them.
• Empty nested list items right below an item's text (- a\n -)
  survive org → md: Markdown read the text over a lone - as a
  heading, losing one item each round trip.
• An org #+OPTIONS: line keeps an author's ^:{} where it is: org →
  md took it out from anywhere, md → org put it back last.
• An indented code line org escapes with a comma (,* x after the
  indentation, in a src or export block) keeps its indentation in
  Markdown: uniorg dropped it with the comma.
• A Markdown code block without a language, or an HTML block, whose
  line org would read as a headline or keyword (* x, #+end_example)
  gets org's comma escape in its example or export block; that line
  ended the block or became a headline. Org → md drops the comma of an
  example block's line too, which uniorg left in (in a CRLF file
  too), but keeps the escapes
  of a block in a drawer or special block, which travels as org text.
• Adjacent Markdown tables stay apart in org: md → org wrote them
  with no blank line between them, which org reads as one table.
• An org table keeps its #+NAME:, #+CAPTION:, #+RESULTS: (an
  empty one too) and other affiliated keywords, and its #+TBLFM:
  formulas, as lines around the Markdown table; both were lost
  without a warning (the keywords by uniorg), in a CRLF file too.
  Several #+TBLFM: lines stay apart: uniorg joined them into one.
• Org links within a file survive a round trip: a link to a headline
  ([[*Some Heading]], or a fuzzy [[Some Heading]]) is a link to the
  heading's GitHub slug (#some-heading), a link to a <<target>> or
  a #+NAME: one to #name, and md → org links them back. They came
  back as file: links to files of that name. A Markdown anchor of a
  heading ([setup](#getting-started)) becomes a heading link, which
  Emacs follows; it was a custom ID link that found nothing.
• An org link to a // path keeps its % as written: org → md
  percent-encoded it, md → org read the scheme-relative url as it is,
  so %25 gained a 25 each round trip.
• Org keyword lines (#+AUTHOR: Jan <jan@x.org>, #+CALL:) go back
  to org as written: md → org read them as Markdown, so an address or
  url became a link nested once more each round trip, and escaped them
  (\n:nil in #+OPTIONS: gained a zero-width space).
• A headline's property values that Markdown may read as syntax (a
  url, *, _) are written below a <!-- morg_properties -->
  comment, and md → org takes the lines below it back as written; it
  read them as Markdown. Unmarked key:: value lines below a heading
  are Markdown, as a person writes them (an Obsidian Dataview field),
  and convert as before.
• An org element written verbatim to Markdown (fixed-width lines, a
  drawer, a special block) with ke...

v0.10.1

Fixed

• Obsidian md → org keeps an embed's size (![[image.png|300]]); it
  became a link description (![[image.png][300]]).
• Obsidian org → md writes an embed's ! unescaped; \![[image.png]]
  was a ! and a link to Obsidian.
• Translating Markdown to or from Logseq md counts a tab in a list
  item's indentation to its tab stop. A fence indented with spaces among
  tab-indented lines drifted on each trip, and a tab-indented item's
  fence could become indented code.

v0.10.0

Removed

• Breaking: the morg normalize subcommand. One format on both
  sides is enough: with one preset it normalizes (--from org and
  --to org, or --input notes.org --output notes.org), with two it
  translates.

Added

• translateOrg(org, { inputPreset, outputPreset }): Logseq org ↔
  Vanilla org within org, changing only the headlines the two write
  differently (an empty block's space, a first line that starts an
  element) and keeping every other line as written (ADR 0006 §4
  amended). Logseq's task markers (NOW, DOING, WAITING, …) are
  declared for Emacs in a #+TODO: line, dropped again on the way
  back; an own #+TODO: line, which Logseq ignores, warns of the
  markers Logseq shows as text. A collapsed block is folded for Emacs
  (:collapsed: true ↔ :VISIBILITY: folded). Emacs syntax Logseq
  misreads ([[*heading]] links, <<<radio>>> targets, keyword lines
  below the first headline, …) stays, with a warning.
• translateMarkdown(md, { inputPreset, outputPreset }): Logseq
  Markdown ↔ Vanilla Markdown within Markdown, blocks ↔ list items and
  headings with their meta as before, page properties ↔ flat
  frontmatter, a block's org blocks ↔ fences, quotes and query code
  blocks; other content stays as written. Obsidian Markdown ↔ Logseq
  Markdown too, an aliased wikilink ↔ a labeled page ref, and
  Obsidian Markdown → Vanilla Markdown: a %%comment%% becomes an
  HTML comment and an inline footnote a footnote (also on the way to
  Logseq Markdown); wikilinks stay, as resolving a note's file needs
  the vault. markdownStyle (CLI --bullet, --emphasis, …) rewrites
  only the markers it names; Logseq Markdown keeps its - bullets.
• Web UI: two dialects of one format translate (Input: Markdown
  (Logseq), Output: Markdown); a translated file is saved under its
  dialect (page.vanilla.md). Changing the input's format still moves
  the output to the other format.
• CLI: one format on both sides with two presets translates
  (morg --input-preset logseq --input page.md --output out.md); it
  was an error.
• Vanilla org → Logseq Markdown warns of [[*heading]] links, which
  Logseq reads as refs to pages of that name.

Changed

• Breaking: Node.js 24 or later (engines.node was >=20, which
  nothing tested; Node 20 is end-of-life).

v0.9.2

Fixed

• Logseq org ↔ Logseq Markdown reads a block's title as inline text: a
  title that looks like org line syntax (: a, - a, 1. a, # a)
  was converted as a fixed-width line, a list or a comment, and a
  : a title holding markup grew with each round trip. A table or a
  rule on the headline line stays one, which Logseq org → Vanilla
  Markdown now does too.

v0.9.1

Fixed

• An empty Markdown heading becomes an org headline with a space after
  its stars, as Emacs needs to read it as one; it was written as bare
  stars, which Emacs and morg read as text.
• Logseq Markdown → Vanilla org writes an empty block as a headline
  with a space after its stars, which Emacs reads as one, instead of
  Logseq org's bare stars.
• Logseq Markdown → Vanilla org writes a block whose first line starts
  a src block, a table, a list or its property drawer with that line
  below an empty headline: on the stars' line, as Logseq org writes it,
  Emacs read it as the title and cut the element off (ADR 0006,
  amended).
• A multi-line HTML comment keeps an empty first or last line as an
  empty org comment line instead of losing it, so such a
  comment comes back as it was.

v0.9.0

Added

• inputPreset and outputPreset conversion options: the dialect the
  input is read in and the one the output is written in, Vanilla when
  left out (ADR 0006). preset still sets both, leaving a side Vanilla
  whose format it has no dialect for; a side preset without one throws,
  as does preset together with another side preset, and normalize*
  with different presets per side.
• --input-preset / --output-preset and inputPreset /
  outputPreset in morg.toml, and the preset name vanilla. Each
  side resolves on its own: side flag, --preset, side key, preset;
  preset next to a different side preset in one place is an error.
• Logseq org → Vanilla Markdown (--input-preset logseq): blocks become
  a nested list, a heading block a heading whose children start a new
  list, numbered blocks an ordered list. Task markers render as Logseq
  shows them (- [ ], - [x], - [ ] LATER …); planning and
  properties become key:: value lines in the item, under the
  orgismKeys names; collapsed and drawers are dropped with a
  warning. A block's title stays inline text, as Logseq reads it, even
  where it looks like org line syntax (: a, - a). After a heading
  block, Markdown has no way back to the top level, so a later
  top-level block reads back as the heading's child. Logseq's inline
  syntax is carried as Logseq Markdown writes it, as text a return trip
  restores: page and block refs
  ([[Page]], [label]([[Page]])), macros ({{video …}}), tags,
  priorities, hiccup, highlights; a query block becomes a query code
  block.
• Vanilla Markdown → Logseq org (--output-preset logseq), as Logseq
  reads a file it did not write: a list item is a block, its nested
  list its children; a heading is a block nesting by level, the text
  below it its content; other text after a list a block of its own.
  Task items, key:: value lines and ordered lists come back as task
  markers, planning, properties and numbered blocks. Reference links
  become inline before the page splits into blocks, and a footnote's
  definition moves to the block of its first reference. Logseq's syntax
  carried in Vanilla Markdown (page and block refs, macros, a query
  code block) is Logseq's again.
• Logseq Markdown ↔ Vanilla org: Vanilla org is read and written as
  Logseq org, which it is, page properties as leading keywords.
• Logseq page properties ↔ plain Vanilla Markdown frontmatter
  (title: P), as Markdown tools read metadata; a keyword that acts in
  Emacs stays in morg_keywords.
• Web UI: an Input and an Output select, each naming a format and its
  dialect (Org, Org (Logseq), Markdown, Markdown (Logseq), Markdown
  (Obsidian)), above their text areas, replace the direction and preset
  selects. The direction follows from the two; the same on both sides
  normalizes, and says so; a dialect change within one format is not
  offered; ⇄ swaps the sides. The demo follows the input; sides
  remembered by 0.8.0 are restored.
• Obsidian ↔ Logseq: an Obsidian wikilink and a Logseq page ref map
  onto each other ([[Page|label]] ↔ [[Page][label]]).

Changed

• Breaking. A FragmentConverter takes a third argument, the
  preset whose syntax a Vanilla side carries. Preset.convertOrg and
  Preset.convertMarkdown get a third argument, the conversion's
  context (exported as ConversionContext): the side the preset is on
  (both, input or output), onWarning and orgismKeys.
• Breaking. Preset holds one dialect per format, each read in
  one direction and written in the other (ADR 0006):
  markdown: { read: { mdast, org }, write } and
  org: { read, write }, typed MarkdownDialect and OrgDialect.
  applyToMdast, applyToUniorg and extractFromUniorg are gone. The
  built-in presets convert as before.

v0.8.0

Added

• Preset.convertOrg(org, convert) and
  Preset.convertMarkdown(markdown, convert), optional hooks that take
  a whole conversion over, for a dialect whose documents are an outline
  of fragments; convert runs the core on one fragment (recording no
  Markdown style: a fragment's style is the preset's).
• The Web UI demo also shows a quote, a nested ordered list and a src
  block switch.
• The Web UI offers a Logseq demo page, in Logseq org or Logseq
  Markdown, while the Logseq preset is selected; an untouched demo
  follows the preset as it follows the direction.

Changed

• Breaking. Logseq preset: a page converts as Logseq's outline of
  blocks. Org headlines ↔ - bullets indented one tab per level, an
  empty block (bare stars) ↔ -, :heading: N ↔ - ## …, a block's
  property drawer ↔ key:: value lines; a block's content converts as
  one fragment, so a code block or table that starts on the headline
  line stays whole. Blocks used to become paragraphs below the nearest
  heading, which flattened every level below it. logseq() takes no
  options anymore (nestUnderHeadings is gone).

Fixed

• A src block's switches and header arguments
  (#+begin_src python -n :results output) travel as the Markdown
  fence's meta (```python -n :results output) instead of being
  dropped both ways.
• A block or drawer kept as org text in Markdown (special, verse,
  center blocks, :LOGBOOK:, …) no longer falls apart on the way back
  when its lines read as Markdown syntax: a blank line followed by an
  indented line became a code block, a # line a heading, a - line a
  list, and
  the round trip did not converge.
• Logseq preset: a block whose content starts with its properties
  (** :PROPERTIES:, as Logseq writes a query block) keeps them as its
  key:: lines instead of growing an :END: line per round trip.
• Logseq preset: a repeated task's :LOGBOOK: state lines take the
  bullet Logseq writes in each format, - State … in org and
  * State … in Markdown.
• Logseq preset: an org plain link (https://…) stays a bare url in
  Markdown, as Logseq writes it, and comes back plain instead of as
  [[https://…]].
• Logseq preset: an email address stays text, written bare in
  Markdown as Logseq does, instead of coming back as a mailto: link.
• Logseq preset: a block that starts with a tag (#meeting) keeps it
  a tag in Markdown; it was written \#meeting, which Logseq reads as
  plain text.
• Logseq preset: an org block written in a Markdown block
  (#+BEGIN_SRC…#+END_SRC, which Logseq reads there too) goes to org
  instead of as escaped text that changed on every round trip; the
  Markdown content of a quote block converts.

v0.7.0

Added

• Logseq preset: page properties map both ways. A page's first block of
  key:: value lines and flat frontmatter entries become the leading
  #+key: value lines Logseq reads as page properties of an org page,
  and come back as a key:: block, values as written. Frontmatter keys
  that act in Emacs (todo, include, setupfile, …) stay inert in
  the frontmatter block. The block used
  to become an invalid headline (* title:: …).
• Preset.applyToMdast(mdast, markdown), an optional hook that runs
  before the generic md → org transform, with the Markdown source at
  hand.

Changed

• Breaking output change. YAML frontmatter no longer becomes org
  keywords (tags: [a] → #+TAGS: a); it travels verbatim in a
  marked comment block, #+begin_comment morg_frontmatter, inert in
  Emacs and in export. Upper-cased keys used to collide with keywords
  that act: #+TAGS declares the tag vocabulary, #+TODO redefines
  the TODO keywords, #+INCLUDE pulls a file into exports (ADR 0005).
  transformMdastToUniorgAst returns the block as a raw text node
  instead of keyword nodes, which transformUniorgAstToMdast reads
  back; a uniorg tree parsed from org text cannot carry the block's
  marker (uniorg drops it), so pass that text as its org option, or
  convert files with convertOrgToMarkdown.
• Breaking output change. An org file's leading keywords
  (#+TITLE, #+STARTUP, …) no longer become plain frontmatter keys;
  they travel as a morg_keywords frontmatter entry and come back as
  keywords, so #+STARTUP keeps its effect. An entry a keyword line
  cannot hold (a line break in the value, a key org does not read as a
  keyword name) stays in the frontmatter block instead.
• Migration: files written by 0.6.0 and earlier carry their frontmatter
  as keywords and now read as org-native keywords (morg_keywords).
  That converges, but Markdown tools no longer see title. Re-convert
  them from the Markdown source, or move the keywords into a
  morg_frontmatter block by hand.

Fixed

• A dual affiliated keyword keeps its second value through Markdown:
  #+CAPTION[short]: long used to come back as a plain #+CAPTION:
  with the short caption glued onto the long one, and
  #+RESULTS[hash]: … as two separate lines.
• An Emacs mode line (# -*- mode: org -*-) stays on the first line
  through Markdown, where Emacs reads it, instead of landing below the
  frontmatter.
• A file-level property drawer (org-roam's :ID:) no longer turns into
  ID:: abc text in Markdown and stays text on the way back, which
  broke org-roam links. It travels as a morg_properties frontmatter
  entry and comes back as the drawer leading the file.
• A leading #+CAPTION:, #+NAME: or other keyword org attaches to
  the element below it (#+SOURCE: is an alias of #+NAME:) no longer
  comes back from Markdown attached to the next paragraph or keyword;
  md → org keeps it apart with a blank line.
• A literal backslash before a letter in Markdown text no longer turns
  into LaTeX or a symbol through org: a path like C:\Users\me used to
  come back as C:$\Users$$\me$, \alpha as α.
• A | in inline code in a Markdown table cell (`a \| b`) no
  longer splits the org table cell: it becomes the lookalike ∣
  (U+2223) in org, reported via onWarning, and \| again in
  Markdown.
• An ordered list starting at 0. keeps its start through org → md
  instead of being renumbered from 1..
• A hard line break inside a list item no longer comes back from org
  with the next line's indentation as literal leading whitespace
  ( ).

v0.6.0

Changed

• Conversion is slower than in 0.5.0: md → org takes about 1.6 to 2.8
  times as long, org → md about 10 % longer. The escapes below have to
  check how org reads the output, which takes org parses of the text
  concerned; org → md guards _. lines and drops the escapes again.
  A performance check in CI now keeps it from growing unnoticed.

Fixed

Mostly found by round-tripping a real Obsidian vault, the rest by code
review and fuzzing; each of these silently changed text.

• Markup touching a word character (`x`s, foo**bar**baz) no
  longer comes back with literal markers. Org only reads markup next to
  whitespace or punctuation, so md → org now separates the two with a
  zero-width space, the org manual's escape character, and org → md
  drops it again.
• Literal text org would read as markup, such as the path /etc/ or an
  escaped \*b\*, stays literal instead of turning italic or bold,
  also when another inline node sits between the markers
  (\*b `x` c\*).
• Relative links and images ([t](notes.md#Some%20Heading)) become
  org file: links ([[file:notes.md::Some Heading][t]]) and come
  back as the same Markdown links. They used to become bare org paths,
  which org reads as a heading search rather than a file, so Emacs
  would not open them and the obsidian preset turned them into
  wikilinks. A path holding a literal %5B, %5D or :: keeps it
  instead of pointing at another file, and an org file: path holding
  # or % ([[file:C# notes.md]]) comes back pointing at that file,
  not at C.
• A heading inside a list item keeps its text as a line of the item
  and reports via onWarning, instead of silently coming back as
  literal \*\* …: org has no headline inside a list.
• A paragraph line that starts like org line syntax, such as an
  escaped 1\., \- or \#, or a lazy continuation line, stays text
  instead of turning into a list item, headline, comment or table. This
  includes a line starting with a footnote reference ([^1] text,
  which org reads as a footnote definition), one following a nested
  list or code block inside a list item, one inside bold or a link
  text spanning two lines, one ending in a bare bullet (1.), and one
  in a logseq block below its first line. The same goes for line
  syntax spanning lines (an escaped \#+begin_src … \#+end_src) or
  depending on the heading above (SCHEDULED: <…>). A paragraph that
  is exactly one of the org constructs morg writes into Markdown
  verbatim (drawers, fixed-width lines, clocks, …) still comes back as
  org.
• A line starting _. or _) converts in both directions: uniorg,
  morg's org parser, took it for a list bullet and failed on the whole
  document, or, when a list followed, silently dropped the line. Org
  reads it as text, and so does morg now.
• List item text after a nested list (- a\n - b\n\n c) stays in
  the outer item instead of moving into the nested one on the next
  round trip: org → md keeps the blank line Markdown needs there.
• A literal \[fn:1] in Markdown text stays text instead of becoming a
  footnote reference.
• Inline code holding a ~ that ends org code early (`a~ b`)
  becomes org =a~ b= instead of breaking the rest of the line. Code
  that a = would end early too stays plain text and reports via
  onWarning.
• #+OPTIONS: ^:nil is honored on org → md: a_b stays a_b
  instead of becoming a\_{b}, and the option stays.
• Bold, italic or strikethrough spanning more than two lines keeps its
  markup: org markup spans at most two lines, so its line endings
  become spaces, as Markdown renders them anyway.
• A heading inside a blockquote keeps its text as a line of the quote
  and reports via onWarning: Emacs ends a quote block at a headline.
• A fenced code block inside a list item keeps its indentation in both
  directions. md → org used to shrink the code's own indentation by up
  to the item's (if x:\n y() lost two spaces), and org → md added
  the item's indentation to every line.
• An escaped \| in a Markdown table cell becomes org's \vert{}
  entity instead of splitting the cell, and comes back as \|.
• Bare underscores and carets (my_notes_2021.md, x^y, also right
  after markup or a link: **a**_b, or in an obsidian wikilink alias:
  [[a_b.md|a_b]]) are no longer read as org sub/superscripts. md → org adds ^:{} to #+OPTIONS: when the text
  needs it; org → md honors it in a top-level #+OPTIONS: and consumes
  it where the text needs it, so an author's own options: ^:{} stays.
• Inline code spanning three or more lines converts to org; its line
  endings (LF, CRLF or CR) become spaces, which is how CommonMark
  renders them anyway.
• Inline code with whitespace at either end, as a padding typo easily
  leaves, converts to org instead of breaking the rest of the
  paragraph: org markup may not start or end with whitespace, so the
  whitespace moves just outside the code. Whitespace-only code stays
  plain text and reports via onWarning.
• An image with an anchor (![a](i.svg#part)) comes back as an image
  instead of a plain link.
• A space between text and markup inside an org table cell is kept;
  only the cell's alignment padding is trimmed.
• With the obsidian preset, an aliased wikilink in a table cell is
  written [[Page\|alias]], as Obsidian does, instead of splitting the
  cell.

v0.5.0

Added

• morg --help, morg -h and a bare morg print usage: the
  normalize command, every flag with its argument and a one-line
  description, and a few worked examples. morg --version prints the
  installed version. A bare invocation previously failed on formats it
  could not determine, which told a first-time reader nothing about
  what the CLI accepts; there was no way to ask at all.

  The help text is rendered from the same table the parser dispatches
  on, so it cannot document a flag that does not exist or omit one that
  does.

Changed

• The site names the npm package and links it. The landing page says
  morg is also a CLI and a library, published as @remigius42/morg,
  and shows the npx line that runs it without an install; both pages
  carry an npm link in the nav beside GitHub. Until now the only route
  from the Web UI to the package was the repository, which documents it
  but does not hand it over.

  The nav had fit 320px exactly, so a fourth item would have put the
  whole page into a horizontal scroll: the nav lists now wrap, which
  costs a second row on the narrowest phones and leaves every wider
  viewport as it was.

Fixed

• An unsupported format names the value it rejected, as in
  Unsupported format 'notes.md', rather than only the two it
  accepts, and a value that has a file extension is answered with the
  likely correction: --from and --to take a format name, and the
  file belongs to --input or --output. They sit beside each other
  and read alike, so a file name in the wrong one is an easy mistake that
  the old message left the reader to find unaided. Failures that come
  from the invocation now also point at morg --help.