diff --git a/src/Parsoid/Poem.php b/src/Parsoid/Poem.php
new file mode 100644
index 0000000..a2dd6b8
--- /dev/null
+++ b/src/Parsoid/Poem.php
@@ -0,0 +1,133 @@
+<?php
+declare( strict_types = 1 );
+
+namespace Wikimedia\Parsoid\Ext\Poem;
+
+use Wikimedia\Parsoid\DOM\DocumentFragment;
+use Wikimedia\Parsoid\Ext\ExtensionModule;
+use Wikimedia\Parsoid\Ext\ExtensionTagHandler;
+use Wikimedia\Parsoid\Ext\ParsoidExtensionAPI;
+use Wikimedia\Parsoid\Ext\PHPUtils;
+use Wikimedia\Parsoid\Utils\DOMCompat;
+
+class Poem extends ExtensionTagHandler implements ExtensionModule {
+
+	/** @inheritDoc */
+	public function getConfig(): array {
+		return [
+			'name' => 'Poem',
+			'domProcessors' => [
+				PoemProcessor::class,
+			],
+			'tags' => [
+				[
+					'name' => 'poem',
+					'handler' => self::class,
+					'options' => [
+						'outputHasCoreMwDomSpecMarkup' => true
+					],
+				]
+			]
+		];
+	}
+
+	public function __construct( ?ParsoidExtensionAPI $extApi = null ) {
+		/* @phan-suppress-previous-line PhanEmptyPublicMethod */
+		/* The dom post processor doesn't need to use $extApi, so ignore it */
+	}
+
+	/** @inheritDoc */
+	public function sourceToDom(
+		ParsoidExtensionAPI $extApi, string $content, array $extArgs
+	): DocumentFragment {
+		/*
+		 * Transform wikitext found in <poem>...</poem>
+		 * 1. Strip leading & trailing newlines
+		 * 2. Suppress indent-pre by replacing leading space with &nbsp;
+		 * 3. Replace colons with <span class='...' style='...'>...</span>
+		 * 4. Add <br/> for newlines except (a) in nowikis (b) after ----
+		 */
+
+		if ( strlen( $content ) > 0 ) {
+			// 1. above
+			$content = PHPUtils::stripPrefix( $content, "\n" );
+			$content = PHPUtils::stripSuffix( $content, "\n" );
+
+			// 2. above
+			$content = preg_replace( '/^ /m', '&nbsp;', $content );
+
+			// 3. above
+			$contentArray = explode( "\n", $content );
+			$contentMap = array_map( static function ( $line ) use ( $extApi ) {
+				$i = 0;
+				$lineLength = strlen( $line );
+				while ( $i < $lineLength && $line[$i] === ':' ) {
+					$i++;
+				}
+				if ( $i > 0 && $i < $lineLength ) {
+					$domFragment = $extApi->htmlToDom( '' );
+					$doc = $domFragment->ownerDocument;
+					$span = $doc->createElement( 'span' );
+					$span->setAttribute( 'class', 'mw-poem-indented' );
+					$span->setAttribute( 'style', 'display: inline-block; margin-inline-start: ' . $i . 'em;' );
+					// $line isn't an HTML text node, it's wikitext that will be passed to extTagToDOM
+					return substr( DOMCompat::getOuterHTML( $span ), 0, -7 ) .
+						ltrim( $line, ':' ) . '</span>';
+				} else {
+					return $line;
+				}
+			}, $contentArray );
+			$content = implode( "\n", $contentMap ); // use faster? preg_replace
+
+			// 4. above
+			// Split on <nowiki>..</nowiki> fragments.
+			// Process newlines inside nowikis in a post-processing pass.
+			// If <br/>s are added here, Parsoid will escape them to plaintext.
+			$splitContent = preg_split( '/(<nowiki>[\s\S]*?<\/nowiki>)/', $content,
+				-1, PREG_SPLIT_DELIM_CAPTURE );
+			$content = implode( '',
+				array_map( static function ( $p, $i ) {
+					if ( $i % 2 === 1 ) {
+						return $p;
+					}
+
+					// This is a hack that exploits the fact that </poem>
+					// cannot show up in the extension's content.
+					return preg_replace( '/^(-+)<\/poem>/m', "\$1\n",
+						preg_replace( '/\n/m', "<br/>\n",
+							preg_replace( '/(^----+)\n/m', '$1</poem>', $p ) ) );
+				},
+				$splitContent,
+				range( 0, count( $splitContent ) - 1 ) )
+			);
+
+		}
+
+		// Add the 'poem' class to the 'class' attribute, or if not found, add it
+		$value = $extApi->findAndUpdateArg( $extArgs, 'class', static function ( string $value ) {
+			return strlen( $value ) ? "poem {$value}" : 'poem';
+		} );
+
+		if ( !$value ) {
+			$extApi->addNewArg( $extArgs, 'class', 'poem' );
+		}
+
+		return $extApi->extTagToDOM( $extArgs, $content, [
+				'wrapperTag' => 'div',
+				'parseOpts' => [ 'extTag' => 'poem' ],
+				// Create new frame, because $content doesn't literally appear in
+				// the parent frame's sourceText (our copy has been munged)
+				'processInNewFrame' => true,
+				// We've shifted the content around quite a bit when we preprocessed
+				// it.  In the future if we wanted to enable selser inside the <poem>
+				// body we should create a proper offset map and then apply it to the
+				// result after the parse, like we do in the Gallery extension.
+				// But for now, since we don't selser the contents, just strip the
+				// DSR info so it doesn't cause problems/confusion with unicode
+				// offset conversion (and so it's clear you can't selser what we're
+				// currently emitting).
+				'clearDSROffsets' => true
+			]
+		);
+	}
+}
diff --git a/src/Parsoid/PoemProcessor.php b/src/Parsoid/PoemProcessor.php
new file mode 100644
index 0000000..3a60da0
--- /dev/null
+++ b/src/Parsoid/PoemProcessor.php
@@ -0,0 +1,75 @@
+<?php
+declare( strict_types = 1 );
+
+namespace Wikimedia\Parsoid\Ext\Poem;
+
+use Wikimedia\Parsoid\DOM\Element;
+use Wikimedia\Parsoid\DOM\Node;
+use Wikimedia\Parsoid\DOM\Text;
+use Wikimedia\Parsoid\Ext\DOMProcessor;
+use Wikimedia\Parsoid\Ext\DOMUtils;
+use Wikimedia\Parsoid\Ext\ParsoidExtensionAPI;
+
+class PoemProcessor extends DOMProcessor {
+
+	/**
+	 * @inheritDoc
+	 */
+	public function wtPostprocess(
+		ParsoidExtensionAPI $extApi, Node $node, array $options
+	): void {
+		$c = $node->firstChild;
+		while ( $c ) {
+			if ( $c instanceof Element ) {
+				if ( DOMUtils::hasTypeOf( $c, 'mw:Extension/poem' ) ) {
+					// Replace newlines found in <nowiki> fragment with <br/>s
+					self::processNowikis( $c );
+				} else {
+					$this->wtPostprocess( $extApi, $c, $options );
+				}
+			}
+			$c = $c->nextSibling;
+		}
+	}
+
+	private function processNowikis( Element $node ): void {
+		$doc = $node->ownerDocument;
+		$c = $node->firstChild;
+		while ( $c ) {
+			if ( !$c instanceof Element ) {
+				$c = $c->nextSibling;
+				continue;
+			}
+
+			if ( !DOMUtils::hasTypeOf( $c, 'mw:Nowiki' ) ) {
+				self::processNowikis( $c );
+				$c = $c->nextSibling;
+				continue;
+			}
+
+			// Replace the nowiki's text node with a combination
+			// of content and <br/>s. Take care to deal with
+			// entities that are still entity-wrapped (!!).
+			$cc = $c->firstChild;
+			while ( $cc ) {
+				$next = $cc->nextSibling;
+				if ( $cc instanceof Text ) {
+					$pieces = preg_split( '/\n/', $cc->nodeValue );
+					$n = count( $pieces );
+					$nl = '';
+					for ( $i = 0;  $i < $n;  $i++ ) {
+						$p = $pieces[$i];
+						$c->insertBefore( $doc->createTextNode( $nl . $p ), $cc );
+						if ( $i < $n - 1 ) {
+							$c->insertBefore( $doc->createElement( 'br' ), $cc );
+							$nl = "\n";
+						}
+					}
+					$c->removeChild( $cc );
+				}
+				$cc = $next;
+			}
+			$c = $c->nextSibling;
+		}
+	}
+}