feat(markdown): support Markdown-It 13, 14, and 15 hosts

.changeset/complete-unicode-paragraphs.md

@@ -0,0 +1,8 @@
+---
+'@bidilens/core': patch
+'@bidilens/react': patch
+'@bidilens/cli': patch
+---
+
+Treat every Unicode 17 `Bidi_Class=Paragraph_Separator` character as a default
+batch and streaming boundary, with matching React, Android, and Rust behavior.

.changeset/finite-option-guards.md

@@ -0,0 +1,7 @@
+---
+'@bidilens/core': patch
+'@bidilens/markdown': patch
+---
+
+Reject non-finite detection and stream options instead of allowing NaN to
+silently disable direction decisions or stream locking.

.changeset/healthy-geese-compare.md

@@ -0,0 +1,14 @@
+---
+'@bidilens/markdown': patch
+---
+
+Expand the optional Markdown-It peer range to cover 13.x, 14.x, and the current
+15.x host line without exposing any parser's types through the public API. The
+adapter now uses a version-neutral structural runtime boundary instead of
+private `markdown-it/lib/*` declarations, and its own source build runs against
+Markdown-It 15. A packed strict TypeScript/peer consumer gate runs all 932
+canonical cases plus host-structure fixtures through all three parser lines.
+Markdown-It 13 and 14 require identical full reports; Markdown-It 15's upstream
+linkifier is allowed its documented HTML difference while BidiLens block,
+security, and isolation reports remain identical. Each host also passes the
+adapter's finalized batch/stream-equivalence check.

.changeset/quiet-marks-detect.md

@@ -0,0 +1,7 @@
+---
+'@bidilens/core': patch
+---
+
+Keep Unicode combining marks attached at isolation boundaries and recognize
+compact technical labels such as `CN X`, `CN IX`, and `HR/BP` without changing
+ordinary natural-language evidence.

.changeset/secure-paragraph-boundaries.md

@@ -0,0 +1,6 @@
+---
+'@bidilens/core': patch
+---
+
+Scope bidi-control balancing to Unicode paragraph boundaries so a PDF or PDI
+cannot close an embedding or isolate opened in an earlier paragraph.

.github/dependabot.yml

@@ -35,8 +35,9 @@ updates:
           - version-update:semver-major
       # markdown-it 15 changes its TypeScript surface: the constructor no
       # longer serves as the instance type and renderer callbacks lose the
-      # contextual types used by our adapter. Adopt that major only alongside
-      # a deliberate compatibility patch and consumer/type-layout validation.
+      # contextual types used by older adapters. BidiLens now uses a
+      # version-neutral structural boundary; keep future major updates paired
+      # with the source-build and packed consumer compatibility gates.
       - dependency-name: "markdown-it"
         update-types:
           - version-update:semver-major
@@ -46,6 +47,13 @@ updates:
     schedule:
       interval: weekly
     open-pull-requests-limit: 3
+    groups:
+      # CodeQL init and analyze must move together. Keeping them in separate
+      # PRs can create a transient configuration mismatch where one action
+      # runs a newer CodeQL version against the other action's older config.
+      codeql-action:
+        patterns:
+          - "github/codeql-action/*"
 
   - package-ecosystem: gradle
     directory: /android

.github/workflows/ci.yml

@@ -97,6 +97,33 @@ jobs:
       - run: pnpm run packages:types
       - run: pnpm run release:check
 
+  markdown-it-compat:
+    name: Markdown-It 13, 14, and 15 compatibility
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          persist-credentials: false
+      - uses: pnpm/action-setup@0977fd99725f1db4007ccb2928dbb4e90d06cc86 # v6.0.10
+        with:
+          version: 10.27.0
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: 22.12.0
+          cache: pnpm
+          cache-dependency-path: pnpm-lock.yaml
+      - run: pnpm install --frozen-lockfile
+      - run: pnpm run markdown-it:compat
+        env:
+          BIDILENS_MARKDOWN_IT_REPORT_DIR: markdown-it-compat-reports
+      - if: ${{ failure() }}
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: markdown-it-compat-reports
+          path: markdown-it-compat-reports
+          if-no-files-found: error
+          retention-days: 7
+
   android:
     name: Android libraries and sample
     runs-on: ubuntu-latest

.github/workflows/release.yml

@@ -30,6 +30,7 @@ jobs:
       - run: pnpm exec playwright install --with-deps chromium firefox webkit
       - run: pnpm run test:visual
       - run: pnpm run packages:types
+      - run: pnpm run markdown-it:compat
       - run: pnpm run deps:audit
       - run: pnpm run release:check
       - run: pnpm run sbom

CONTRIBUTING.md

@@ -84,6 +84,7 @@ Changesets entry for a public API or behavior change. Before a release run:
 pnpm run check
 pnpm run test:visual
 pnpm run packages:types
+pnpm run markdown-it:compat
 pnpm run deps:audit
 pnpm run release:check
 pnpm run sbom

IMPACT.md

@@ -7,13 +7,15 @@
 - 932 schema-validated direction fixtures with numbered logical words;
 - 0 fixtures currently certified by a native-language reviewer;
 - property-based stream/source/range checks;
-- 396 unit/property/action tests with 94.80% overall and 95.81% core line coverage,
-  including multilingual strict-security false-positive regressions;
-- 24 browser/visual checks spanning Chromium, Firefox, and WebKit, including
+- 424 unit/property/action tests with 92.23% statements, 86.23% branches,
+  94.84% functions, and 95.05% lines overall (96.34% core lines), including
+  paragraph-boundary security, isolate balancing, and non-finite-option
+  regressions;
+- 30 Playwright browser/visual tests spanning Chromium, Firefox, and WebKit, including
   structured Markdown, real Chromium clipboard verification, and no-build
   standalone Web Component loading, plus the full offline bilingual playground
   controls, corpus, copy invariant, theme, and export flow;
-- a self-contained 196,766-byte Node 24 GitHub Action bundle with source tests
+- a self-contained 199,733-byte Node 24 GitHub Action bundle with source tests
   and built-artifact safe/strict-failure probes;
 - reproducible Unicode 17.0.0 source and generated tables;
 - clean tarball installation, strict consumer type-check, runtime imports, and
@@ -28,9 +30,12 @@ the [build report](docs/V1_BUILD_REPORT.md).
 
 ## Not yet measured
 
-There are no claimed users, downloads, stars, external contributors, merged
-upstream patches, production deployments, accessibility lab results, security
-audit, grants, or sponsorships. Those outcomes require real external evidence.
+There are no claimed users, downloads, stars, external contributors, production
+deployments, accessibility lab results, independent security audit, grants, or
+sponsorships. One focused native implementation based on the BidiLens policy
+merged upstream in Streamdown; that is integration evidence, not proof of
+BidiLens dependency adoption or continued production use. All other outcomes
+require real external evidence.
 
 ## Intended impact
 

README.md

@@ -74,6 +74,10 @@ the immutable logical source, and isolates the technical identifier.
 |---|---|
 | ![The flagship sentence rendered with the incorrect automatic LTR base](tests/visual/__screenshots__/chromium/flagship-auto.png) | ![The same logical sentence rendered with the correct RTL base and isolated React token](tests/visual/__screenshots__/chromium/flagship-toolkit.png) |
 
+The medical Markdown, abbreviation, combining-mark, quote, list, and table
+cases transcribed from mixed AI output are documented in the
+[screenshot regression cases](docs/SCREENSHOT_CASES.md).
+
 ## Why this is more than an RTL stylesheet
 
 | Common approach | Limitation | BidiLens behavior |
@@ -146,7 +150,7 @@ reordering and shaping; BidiLens supplies the application structure they need.
 | `@bidilens/core` | Analysis, evidence, isolation, streaming, and security |
 | `@bidilens/dom` | DOM annotation, restoration, CSS policy, and observation |
 | `@bidilens/html` | XSS-safe semantic HTML serialization |
-| `@bidilens/markdown` | unified/remark/rehype and Markdown-It plugins plus rich Markdown-It streaming |
+| `@bidilens/markdown` | unified/remark/rehype and Markdown-It 13/14/15 plugins plus rich Markdown-It streaming |
 | `@bidilens/playwright` | Rendering, isolation, selection, clipboard, and geometry assertions |
 | `@bidilens/react` | SSR-safe components and streaming hooks |
 | `@bidilens/spec` | Versioned cross-language JSON Schemas and schema registry |
@@ -289,7 +293,8 @@ cargo clippy --manifest-path rust/Cargo.toml --all-targets --all-features -- -D
 ```
 
 `verify:production` runs the complete quality, three-browser visual, package
-type, dependency-audit, SBOM, and clean-release sequence. `release:check`
+type, packed Markdown-It 13/14/15 compatibility, dependency-audit, SBOM, and
+clean-release sequence. `release:check`
 builds and packs every public package, inspects tarball contents, installs all
 tarballs into a temporary consumer, compiles with `skipLibCheck: false`, and
 executes adapter/runtime assertions—including the pure-LTR non-interference

action/dist/index.cjs

@@ -6047,6 +6047,18 @@ function classifyCharacter(character) {
   return isInRanges(codePoint, NATURAL_LETTER_RANGES) ? classifyBidiStrongCharacter(character) : "neutral";
 }
 
+// packages/core/src/options.ts
+function boundedNumberOption(name, value, defaultValue, minimum, maximum) {
+  const resolved = value ?? defaultValue;
+  if (!Number.isFinite(resolved)) {
+    throw new RangeError(`${name} must be a finite number.`);
+  }
+  return Math.min(maximum, Math.max(minimum, resolved));
+}
+
+// packages/core/src/paragraph.ts
+var DEFAULT_PARAGRAPH_SEPARATOR_SOURCE = "\\r\\n|\\n|\\r|\\u0085|[\\u001C-\\u001E]|\\u2029";
+
 // packages/core/src/detect.ts
 var DEFAULT_OPTIONS = {
   strategy: "content-majority",
@@ -6127,8 +6139,20 @@ function normalizeOptions(options = {}) {
     strategy,
     fallback: options.fallback ?? options.inheritedDirection ?? DEFAULT_OPTIONS.fallback,
     inheritedDirection: options.inheritedDirection ?? DEFAULT_OPTIONS.inheritedDirection,
-    minimumStrongCharacters: Math.max(1, options.minimumStrongCharacters ?? DEFAULT_OPTIONS.minimumStrongCharacters),
-    majorityThreshold: Math.min(1, Math.max(0.5, options.majorityThreshold ?? DEFAULT_OPTIONS.majorityThreshold)),
+    minimumStrongCharacters: boundedNumberOption(
+      "minimumStrongCharacters",
+      options.minimumStrongCharacters,
+      DEFAULT_OPTIONS.minimumStrongCharacters,
+      1,
+      Number.POSITIVE_INFINITY
+    ),
+    majorityThreshold: boundedNumberOption(
+      "majorityThreshold",
+      options.majorityThreshold,
+      DEFAULT_OPTIONS.majorityThreshold,
+      0.5,
+      1
+    ),
     // Compatibility/strict first-strong modes must see the real first strong
     // character (including a leading technical identifier), like dir="auto".
     excludeTechnicalTokens: options.excludeTechnicalTokens ?? majorityStrategy,
@@ -6403,6 +6427,9 @@ function findTechnicalTokenRanges(text, technicalIdentifiers = []) {
   addMatches(text, ranges, /\bv?\d+(?:\.\d+){1,}\b/gu, "version");
   addMatches(text, ranges, /\b[0-9a-f]{7,40}\b/giu, "hash");
   addMatches(text, ranges, /(?<![\p{L}\p{N}_])[+-]?(?:\d+(?:[.,]\d+)?|[\u0660-\u0669]+(?:[\u066B\u066C][\u0660-\u0669]+)?|[\u06F0-\u06F9]+(?:[.,][\u06F0-\u06F9]+)?)(?![\p{L}\p{N}_])/gu, "number");
+  addMatches(text, ranges, /\b[A-Z]{1,4}\s+(?:[IVXLCDM]{1,8}|\d{1,3})\b/gu, "identifier");
+  addMatches(text, ranges, /\b[A-Z]{1,4}\/[A-Z]{1,4}\b/gu, "identifier");
+  addMatches(text, ranges, /\b[A-Z]\b(?=\s*(?:=|:|→|->))/gu, "identifier");
   const words = /\b[A-Za-z][A-Za-z0-9_.-]*\b/gu;
   const customIdentifiers = customTechnicalIdentifiers(technicalIdentifiers);
   const uppercaseProse = usesUppercaseProse(text);
@@ -6478,7 +6505,7 @@ function firstBidiStrongCharacter(text) {
 }
 function splitParagraphs(text) {
   const paragraphs = [];
-  const separator = /\r\n|\n|\r|\u2029/gu;
+  const separator = new RegExp(DEFAULT_PARAGRAPH_SEPARATOR_SOURCE, "gu");
   let start = 0;
   let match;
   while ((match = separator.exec(text)) !== null) {
@@ -6550,6 +6577,10 @@ function analyzeText(text, options = {}) {
 }
 
 // packages/core/src/security.ts
+var UAX9_PARAGRAPH_SEPARATOR = new RegExp(
+  `\\r\\n|\\n|\\r|\\u0085|[${String.fromCodePoint(28)}-${String.fromCodePoint(30)}]|\\u2029`,
+  "gu"
+);
 var CONTROL_METADATA = /* @__PURE__ */ new Map([
   [1564, { name: "ARABIC LETTER MARK", risk: "low", category: "mark" }],
   [8206, { name: "LEFT-TO-RIGHT MARK", risk: "low", category: "mark" }],
@@ -6638,7 +6669,7 @@ function lastFrameIndex(stack, kind) {
   }
   return -1;
 }
-function balanceFindings(controls) {
+function balanceParagraph(controls, boundary) {
   const findings = [];
   const stack = [];
   for (const control of controls) {
@@ -6699,14 +6730,28 @@ function balanceFindings(controls) {
     findings.push({
       code: frame.kind === "isolate" ? "BIDI_UNCLOSED_ISOLATE" : "BIDI_UNCLOSED_EMBEDDING",
       severity: "high",
-      message: `${frame.control.name} is not terminated before the end of the text.`,
+      message: `${frame.control.name} is not terminated before ${boundary === "paragraph" ? "the paragraph boundary" : "the end of the text"}.`,
       sourceRange: rangeFor(frame.control),
       remediation: frame.kind === "isolate" ? "Add the matching PDI or remove the isolate opener." : "Add the matching PDF or remove the embedding/override opener.",
       control: frame.control
     });
   }
   return findings;
 }
+function balanceFindings(text, controls) {
+  const findings = [];
+  let controlIndex = 0;
+  for (const match of text.matchAll(UAX9_PARAGRAPH_SEPARATOR)) {
+    const paragraphControls = [];
+    while (controlIndex < controls.length && controls[controlIndex].index < match.index) {
+      paragraphControls.push(controls[controlIndex]);
+      controlIndex += 1;
+    }
+    findings.push(...balanceParagraph(paragraphControls, "paragraph"));
+  }
+  findings.push(...balanceParagraph(controls.slice(controlIndex), "text"));
+  return findings;
+}
 function isAsciiIdentifierCharacter(value) {
   return value !== void 0 && /^[A-Za-z0-9_$]$/u.test(value);
 }
@@ -6767,7 +6812,7 @@ function scanBidiSecurity(text, options = {}) {
   const controls = findBidiControls(text);
   const findings = [
     ...controls.map(controlFinding),
-    ...balanceFindings(controls),
+    ...balanceFindings(text, controls),
     ...invisibleCharacterFindings(text)
   ].sort((a, b) => a.sourceRange.utf16.start - b.sourceRange.utf16.start || a.code.localeCompare(b.code));
   const hasHigh = findings.some((finding) => finding.severity === "high");
@@ -6904,12 +6949,12 @@ function mergeAdjacent(runs) {
 function trimNeutralBoundaries(text, start, end) {
   while (start < end) {
     const character = text.slice(start).match(/^./su)?.[0];
-    if (!character || classifyCharacter(character) !== "neutral") break;
+    if (!character || classifyCharacter(character) !== "neutral" || new RegExp("^\\p{M}$", "u").test(character)) break;
     start += character.length;
   }
   while (end > start) {
     const character = text.slice(0, end).match(/.$/su)?.[0];
-    if (!character || classifyCharacter(character) !== "neutral") break;
+    if (!character || classifyCharacter(character) !== "neutral" || new RegExp("^\\p{M}$", "u").test(character)) break;
     end -= character.length;
   }
   return { start, end };
@@ -7437,7 +7482,7 @@ function highestFindingRisk(findings) {
 function sourcePosition(text, utf16Offset) {
   let lineNumber = 1;
   let lineStart = 0;
-  const newline = /\r\n|\n|\r/gu;
+  const newline = new RegExp(`${DEFAULT_PARAGRAPH_SEPARATOR_SOURCE}|\\u2028`, "gu");
   let match;
   while ((match = newline.exec(text)) !== null && match.index < utf16Offset) {
     lineNumber += 1;

android/README.md

@@ -176,7 +176,7 @@ pnpm run android:check
 Current executable evidence includes:
 
 - all 932 canonical direction fixtures and declared isolation plans in Kotlin;
-- 23 core, 9 Views/Robolectric, and 8 Compose JVM tests;
+- 29 core, 11 Views/Robolectric, and 9 Compose JVM tests;
 - 3 Views and 3 Compose UI tests on an Android 16/API 36.1 emulator;
 - release AAR assembly, sample APK assembly, and Android lint;
 - an isolated consumer build against the generated Maven-local coordinates;

android/core/src/main/kotlin/io/github/codeinscrubs/bidilens/core/BidiAnalyzer.kt

@@ -121,10 +121,12 @@ private fun analyzeParagraph(
     )
 }
 
+private val defaultParagraphSeparator = Regex("\\r\\n|\\n|\\r|\\u0085|[\\u001C-\\u001E]|\\u2029")
+
 private fun splitParagraphs(text: String): List<Pair<String, Int>> {
     val result = mutableListOf<Pair<String, Int>>()
     var start = 0
-    for (match in Regex("\\r\\n|\\n|\\r|\\u2029").findAll(text)) {
+    for (match in defaultParagraphSeparator.findAll(text)) {
         result += text.substring(start, match.range.first) to start
         start = match.range.last + 1
     }