From 6d83a2eb21d49c2fc14235100c327cd19d6e46c4 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 12 Nov 2026 01:12:02 +0000
Subject: [PATCH 1/4] Initial plan
From 02902f3601dd3f88ce0ead0b4e3878bedcd31e98 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 12 Nov 2026 01:21:53 +0000
Subject: [PATCH 2/4] Handle quote-adjacent HTML attributes
Co-authored-by: lpcox <15877973+lpcox@users.noreply.github.com>
---
.../js/sanitize_content_attributes.test.cjs | 30 +++
actions/setup/js/sanitize_content_core.cjs | 171 ++++++++++++------
2 files changed, 150 insertions(+), 51 deletions(-)
create mode 100644 actions/setup/js/sanitize_content_attributes.test.cjs
diff --git a/actions/setup/js/sanitize_content_attributes.test.cjs b/actions/setup/js/sanitize_content_attributes.test.cjs
new file mode 100644
index 00000000000..23f1a9a4072
--- /dev/null
+++ b/actions/setup/js/sanitize_content_attributes.test.cjs
@@ -0,0 +1,30 @@
+import { afterEach, beforeEach, describe, expect, it, vi } from "vitest";
+import { createRequire } from "node:module";
+
+const require = createRequire(import.meta.url);
+const { convertXmlTags, sanitizeContentCore, clearRedactedDomains } = require("./sanitize_content_core.cjs");
+
+beforeEach(() => {
+ vi.stubGlobal("core", { info: vi.fn(), debug: vi.fn(), warning: vi.fn(), error: vi.fn() });
+});
+
+afterEach(() => {
+ clearRedactedDomains();
+ vi.unstubAllGlobals();
+});
+
+describe("quote-aware HTML attribute sanitization", () => {
+ it.each([
+ ['
', '
'],
+ ["
", "
"],
+ ['
', '
'],
+ ['ok', 'ok'],
+ ['ok', 'ok'],
+ [']]>', '(![CDATA[(b title="left > right")]])'],
+ ])("converts %j consistently", (input, expected) => {
+ expect(convertXmlTags(input)).toBe(expected);
+ expect(sanitizeContentCore(input)).toBe(expected);
+ });
+});
diff --git a/actions/setup/js/sanitize_content_core.cjs b/actions/setup/js/sanitize_content_core.cjs
index 83489f7fefb..b14578f3867 100644
--- a/actions/setup/js/sanitize_content_core.cjs
+++ b/actions/setup/js/sanitize_content_core.cjs
@@ -1226,65 +1226,134 @@ function convertXmlTags(s) {
return /^https:\/\/[\w.-]+(?::\d+)?(\/[^\s<>|]*)?(?:\|[^<>]*)?$/.test(tagContent);
}
- // First, process CDATA sections specially - convert tags inside them and the CDATA markers
- s = s.replace(//g, (match, content) => {
- // Convert tags inside CDATA content
- const convertedContent = content.replace(/<(\/?[A-Za-z][A-Za-z0-9]*(?:[^>]*?))>/g, "($1)");
- // Return with CDATA markers also converted to parentheses
- return `(![CDATA[${convertedContent}]])`;
- });
-
/**
- * Strips dangerous HTML attributes from an allowed tag's content string.
- * Removes on* event handler attributes (e.g. onclick, ontoggle), style,
- * title, and data-* attributes in all quoting forms (double-quoted,
- * single-quoted, unquoted, bare).
- *
- * title= and data-* are stripped because their values are invisible in
- * GitHub's rendered markdown (title= appears only as a hover-tooltip;
- * data-* attributes are stripped by GFM's sanitizer) but arrive at the
- * agent verbatim in raw text, creating a steganographic injection channel
- * structurally identical to the one fixed for markdown link titles.
- *
- * Safe attributes such as class, open, lang, id, etc. are preserved.
- *
- * Note: `\s+` (requiring at least one whitespace before the attribute name) is
- * intentional — HTML attributes are always separated from the tag name and from
- * each other by at least one whitespace character. Using `\s*` would risk false
- * matches inside tag names (e.g. matching "ong" inside "strong").
- *
- * @param {string} tagContent - Tag content without surrounding angle brackets
- * @returns {string} Tag content with dangerous attributes removed
+ * Finds a tag's closing angle bracket without treating quoted values as boundaries.
+ * An unterminated quote falls back to its first `>` so malformed markup is neutralized.
+ * @param {string} source
+ * @param {number} start
+ * @returns {{end: number, malformed: boolean}}
*/
- function stripDangerousAttributes(tagContent) {
- // Match: one-or-more whitespace-or-slash + (on* | style | title | data-*) + optional =value
- // Value forms: "...", '...', or unquoted (no whitespace / > / quote chars), or bare (no =)
- // The unquoted form excludes >, whitespace, and all quote characters (', ", `) so it
- // cannot consume the closing > of the tag or straddle other attribute values.
- // Using [\s/]+ (instead of \s+) also strips dangerous attributes that are immediately
- // preceded by a "/" with no space — e.g. the malformed
.
- return tagContent.replace(/[\s/]+(?:on\w+|style|title|data-[\w-]+)(?:\s*=\s*(?:"[^"]*"|'[^']*'|[^\s>"'`]*))?/gi, "");
+ function findTagEnd(source, start) {
+ let quote = "";
+ let firstQuotedGreaterThan = -1;
+ for (let index = start + 1; index < source.length; index++) {
+ const character = source[index];
+ if (quote) {
+ if (character === quote) {
+ quote = "";
+ } else if (character === ">" && firstQuotedGreaterThan < 0) {
+ firstQuotedGreaterThan = index;
+ }
+ } else if (character === '"' || character === "'") {
+ quote = character;
+ } else if (character === ">") {
+ return { end: index, malformed: false };
+ }
+ }
+ return quote && firstQuotedGreaterThan >= 0 ? { end: firstQuotedGreaterThan, malformed: true } : { end: -1, malformed: false };
}
- // Convert opening tags: or to (tag) or (tag attr="value")
- // Convert closing tags: to (/tag)
- // Convert self-closing tags: or to (tag/) or (tag /)
- // But preserve allowed safe tags (with dangerous attributes stripped)
- return s.replace(/<(\/?[A-Za-z!][^>]*?)>/g, (match, tagContent) => {
- if (isHttpsAngleBracketAutolink(tagContent)) {
- return match;
+ /**
+ * Applies a transformation to tag-like markup while respecting quoted values.
+ * @param {string} source
+ * @param {(tagContent: string, match: string, malformed: boolean) => string} transform
+ * @returns {string}
+ */
+ function replaceTagMarkup(source, transform) {
+ let output = "";
+ let written = 0;
+ for (let index = 0; index < source.length; index++) {
+ if (source[index] !== "<") continue;
+ const firstCharacter = source[index + 1] === "/" ? source[index + 2] : source[index + 1];
+ if (!firstCharacter || !/[A-Za-z!]/.test(firstCharacter)) continue;
+
+ const boundary = findTagEnd(source, index);
+ if (boundary.end < 0) return written ? output + source.slice(written) : source;
+ const end = boundary.end + 1;
+ const match = source.slice(index, end);
+ const tagContent = source.slice(index + 1, boundary.end);
+ output += source.slice(written, index) + transform(tagContent, match, boundary.malformed);
+ written = end;
+ index = boundary.end;
}
- // Extract tag name from the content (handle closing tags and attributes)
+ return written ? output + source.slice(written) : source;
+ }
+
+ /**
+ * Strips on* event handlers, style, title, and data-* attributes from tag content.
+ * Attribute values are consumed as quoted or unquoted tokens before the next
+ * attribute is considered, including when a quoted value is immediately followed
+ * by another attribute without whitespace.
+ * @param {string} tagContent
+ * @returns {string}
+ */
+ function stripDangerousAttributes(tagContent) {
const tagNameMatch = tagContent.match(/^\/?\s*([A-Za-z][A-Za-z0-9]*)/);
- if (tagNameMatch) {
- const tagName = tagNameMatch[1].toLowerCase();
- if (allowedTags.includes(tagName)) {
- // Strip dangerous attributes (on* event handlers, style, title, data-*) before preserving
- const sanitizedContent = stripDangerousAttributes(tagContent);
- return `<${sanitizedContent}>`;
+ if (!tagNameMatch) return tagContent;
+
+ let position = tagNameMatch[0].length;
+ let written = 0;
+ let output = "";
+ while (position < tagContent.length) {
+ const separatorStart = position;
+ while (position < tagContent.length && (/\s/.test(tagContent[position]) || tagContent[position] === "/")) {
+ position++;
+ }
+ if (position >= tagContent.length) break;
+
+ const nameStart = position;
+ while (position < tagContent.length && !/[\s/=>"'`<]/.test(tagContent[position])) {
+ position++;
+ }
+ if (position === nameStart) {
+ position++;
+ continue;
+ }
+ const attributeName = tagContent.slice(nameStart, position);
+ let attributeEnd = position;
+ let valueStart = position;
+ while (valueStart < tagContent.length && /\s/.test(tagContent[valueStart])) valueStart++;
+ if (tagContent[valueStart] === "=") {
+ valueStart++;
+ while (valueStart < tagContent.length && /\s/.test(tagContent[valueStart])) valueStart++;
+ if (tagContent[valueStart] === '"' || tagContent[valueStart] === "'") {
+ const quote = tagContent[valueStart];
+ const closingQuote = tagContent.indexOf(quote, valueStart + 1);
+ attributeEnd = closingQuote < 0 ? tagContent.length : closingQuote + 1;
+ } else {
+ attributeEnd = valueStart;
+ while (attributeEnd < tagContent.length && !/[\s>"'`<]/.test(tagContent[attributeEnd])) {
+ attributeEnd++;
+ }
+ }
+ }
+
+ if (/^(?:on\w+|style|title|data-[\w-]+)$/i.test(attributeName)) {
+ output += tagContent.slice(written, separatorStart);
+ written = attributeEnd;
}
+ position = Math.max(attributeEnd, position);
+ }
+ return written ? output + tagContent.slice(written) : tagContent;
+ }
+
+ // Convert tags inside CDATA sections, then convert the CDATA markers themselves.
+ s = s.replace(//g, (_match, content) => {
+ const convertedContent = replaceTagMarkup(content, tagContent => `(${tagContent})`);
+ return `(![CDATA[${convertedContent}]])`;
+ });
+
+ // Convert opening, closing, and self-closing tags while retaining only allowed tags.
+ // Malformed quoted markup is always neutralized rather than repaired into an allowed tag.
+ return replaceTagMarkup(s, (tagContent, match, malformed) => {
+ if (malformed) return `(${tagContent})`;
+ if (isHttpsAngleBracketAutolink(tagContent)) return match;
+
+ const tagNameMatch = tagContent.match(/^\/?\s*([A-Za-z][A-Za-z0-9]*)/);
+ if (tagNameMatch && allowedTags.includes(tagNameMatch[1].toLowerCase())) {
+ return `<${stripDangerousAttributes(tagContent)}>`;
}
- return `(${tagContent})`; // Convert other tags to parentheses
+ return `(${tagContent})`;
});
}
From f7b4550917eb0a1cd9a84ebc44467db1b868931b Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 12 Nov 2026 02:18:41 +0000
Subject: [PATCH 3/4] Fix sanitizer parser review findings
Co-authored-by: lpcox <15877973+lpcox@users.noreply.github.com>
---
.../js/sanitize_content_attributes.test.cjs | 17 ++-
actions/setup/js/sanitize_content_core.cjs | 108 ++++++++++++------
2 files changed, 87 insertions(+), 38 deletions(-)
diff --git a/actions/setup/js/sanitize_content_attributes.test.cjs b/actions/setup/js/sanitize_content_attributes.test.cjs
index 23f1a9a4072..32b14c6032f 100644
--- a/actions/setup/js/sanitize_content_attributes.test.cjs
+++ b/actions/setup/js/sanitize_content_attributes.test.cjs
@@ -20,11 +20,26 @@ describe("quote-aware HTML attribute sanitization", () => {
['
', '
'],
['ok', 'ok'],
['ok', 'ok'],
+ ['ok', 'ok'],
+ ['ok', 'ok'],
[']]>', '(![CDATA[(b title="left > right")]])'],
+ [' right">]]>', '(![CDATA[(b title="left > right")]])'],
+ ["", ""],
+ ['', '(wrapper value="<system>payload</system>")'],
+ ['payload">]]>', '(![CDATA[(wrapper value="<system>payload</system>")]])'],
+ ['', '(b"x=" onmouseover=alert(1)//"")'],
+ [''],
])("converts %j consistently", (input, expected) => {
expect(convertXmlTags(input)).toBe(expected);
expect(sanitizeContentCore(input)).toBe(expected);
});
+
+ it("keeps scanning after unterminated tags without quadratic rescanning", () => {
+ const input = `|]*)?(?:\|[^<>]*)?$/.test(tagContent);
}
- /**
- * Finds a tag's closing angle bracket without treating quoted values as boundaries.
- * An unterminated quote falls back to its first `>` so malformed markup is neutralized.
- * @param {string} source
- * @param {number} start
- * @returns {{end: number, malformed: boolean}}
- */
- function findTagEnd(source, start) {
- let quote = "";
- let firstQuotedGreaterThan = -1;
- for (let index = start + 1; index < source.length; index++) {
+ function createTagScanner(source) {
+ const length = source.length;
+ const normal = new Int32Array(length + 1);
+ const expectValue = new Int32Array(length + 1);
+ const unquotedValue = new Int32Array(length + 1);
+ const singleQuotedValue = new Int32Array(length + 1);
+ const doubleQuotedValue = new Int32Array(length + 1);
+ const nextGreaterThan = new Int32Array(length + 1);
+ nextGreaterThan[length] = -1;
+
+ for (let index = length - 1; index >= 0; index--) {
const character = source[index];
- if (quote) {
- if (character === quote) {
- quote = "";
- } else if (character === ">" && firstQuotedGreaterThan < 0) {
- firstQuotedGreaterThan = index;
- }
- } else if (character === '"' || character === "'") {
- quote = character;
+ nextGreaterThan[index] = character === ">" ? index : nextGreaterThan[index + 1];
+
+ normal[index] = character === ">" ? index + 1 : character === "=" ? expectValue[index + 1] : normal[index + 1];
+ if (/\s/.test(character)) {
+ expectValue[index] = expectValue[index + 1];
+ unquotedValue[index] = normal[index + 1];
} else if (character === ">") {
- return { end: index, malformed: false };
+ expectValue[index] = index + 1;
+ unquotedValue[index] = index + 1;
+ } else {
+ expectValue[index] = character === "'" ? singleQuotedValue[index + 1] : character === '"' ? doubleQuotedValue[index + 1] : unquotedValue[index + 1];
+ unquotedValue[index] = unquotedValue[index + 1];
}
+
+ singleQuotedValue[index] = character === "'" ? normal[index + 1] : singleQuotedValue[index + 1] > 0 ? singleQuotedValue[index + 1] : character === ">" ? -(index + 1) : singleQuotedValue[index + 1];
+ doubleQuotedValue[index] = character === '"' ? normal[index + 1] : doubleQuotedValue[index + 1] > 0 ? doubleQuotedValue[index + 1] : character === ">" ? -(index + 1) : doubleQuotedValue[index + 1];
}
- return quote && firstQuotedGreaterThan >= 0 ? { end: firstQuotedGreaterThan, malformed: true } : { end: -1, malformed: false };
+
+ return {
+ nextGreaterThan,
+ findTagEnd(start) {
+ const result = normal[start + 1];
+ return result > 0 ? { end: result - 1, malformed: false } : result < 0 ? { end: -result - 1, malformed: true } : { end: -1, malformed: false };
+ },
+ };
}
/**
* Applies a transformation to tag-like markup while respecting quoted values.
* @param {string} source
* @param {(tagContent: string, match: string, malformed: boolean) => string} transform
+ * @param {boolean} [preserveAutolinks]
* @returns {string}
*/
- function replaceTagMarkup(source, transform) {
+ function replaceTagMarkup(source, transform, preserveAutolinks = false) {
+ const scanner = createTagScanner(source);
let output = "";
let written = 0;
for (let index = 0; index < source.length; index++) {
@@ -1267,8 +1281,20 @@ function convertXmlTags(s) {
const firstCharacter = source[index + 1] === "/" ? source[index + 2] : source[index + 1];
if (!firstCharacter || !/[A-Za-z!]/.test(firstCharacter)) continue;
- const boundary = findTagEnd(source, index);
- if (boundary.end < 0) return written ? output + source.slice(written) : source;
+ const firstGreaterThan = scanner.nextGreaterThan[index + 1];
+ if (preserveAutolinks && firstGreaterThan >= 0 && isHttpsAngleBracketAutolink(source.slice(index + 1, firstGreaterThan))) {
+ output += source.slice(written, firstGreaterThan + 1);
+ written = firstGreaterThan + 1;
+ index = firstGreaterThan;
+ continue;
+ }
+
+ const boundary = scanner.findTagEnd(index);
+ if (boundary.end < 0) {
+ output += source.slice(written, index + 1);
+ written = index + 1;
+ continue;
+ }
const end = boundary.end + 1;
const match = source.slice(index, end);
const tagContent = source.slice(index + 1, boundary.end);
@@ -1279,6 +1305,10 @@ function convertXmlTags(s) {
return written ? output + source.slice(written) : source;
}
+ function escapeAngleBrackets(text) {
+ return text.replace(/[<>]/g, character => (character === "<" ? "<" : ">"));
+ }
+
/**
* Strips on* event handlers, style, title, and data-* attributes from tag content.
* Attribute values are consumed as quoted or unquoted tokens before the next
@@ -1322,13 +1352,13 @@ function convertXmlTags(s) {
attributeEnd = closingQuote < 0 ? tagContent.length : closingQuote + 1;
} else {
attributeEnd = valueStart;
- while (attributeEnd < tagContent.length && !/[\s>"'`<]/.test(tagContent[attributeEnd])) {
+ while (attributeEnd < tagContent.length && !/\s/.test(tagContent[attributeEnd])) {
attributeEnd++;
}
}
}
- if (/^(?:on\w+|style|title|data-[\w-]+)$/i.test(attributeName)) {
+ if (/^(?:on\w+|style$|title$|data-)/i.test(attributeName)) {
output += tagContent.slice(written, separatorStart);
written = attributeEnd;
}
@@ -1339,22 +1369,26 @@ function convertXmlTags(s) {
// Convert tags inside CDATA sections, then convert the CDATA markers themselves.
s = s.replace(//g, (_match, content) => {
- const convertedContent = replaceTagMarkup(content, tagContent => `(${tagContent})`);
+ const convertedContent = replaceTagMarkup(content, tagContent => `(${escapeAngleBrackets(tagContent)})`);
return `(![CDATA[${convertedContent}]])`;
});
// Convert opening, closing, and self-closing tags while retaining only allowed tags.
// Malformed quoted markup is always neutralized rather than repaired into an allowed tag.
- return replaceTagMarkup(s, (tagContent, match, malformed) => {
- if (malformed) return `(${tagContent})`;
- if (isHttpsAngleBracketAutolink(tagContent)) return match;
-
- const tagNameMatch = tagContent.match(/^\/?\s*([A-Za-z][A-Za-z0-9]*)/);
- if (tagNameMatch && allowedTags.includes(tagNameMatch[1].toLowerCase())) {
- return `<${stripDangerousAttributes(tagContent)}>`;
- }
- return `(${tagContent})`;
- });
+ return replaceTagMarkup(
+ s,
+ (tagContent, match, malformed) => {
+ if (malformed) return `(${escapeAngleBrackets(tagContent)})`;
+
+ const tagNameMatch = tagContent.match(/^\/?\s*([A-Za-z][A-Za-z0-9]*)/);
+ const tagNameDelimiter = tagNameMatch && tagContent[tagNameMatch[0].length];
+ if (tagNameMatch && (tagNameDelimiter === undefined || /[\s/]/.test(tagNameDelimiter)) && allowedTags.includes(tagNameMatch[1].toLowerCase())) {
+ return `<${stripDangerousAttributes(tagContent)}>`;
+ }
+ return `(${escapeAngleBrackets(tagContent)})`;
+ },
+ true
+ );
}
/**
From 2e05c4b6629be9179fdbf78da90dea8af04f0e95 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Thu, 12 Nov 2026 02:20:05 +0000
Subject: [PATCH 4/4] Fix sanitizer delimiter nullability
Co-authored-by: lpcox <15877973+lpcox@users.noreply.github.com>
---
actions/setup/js/sanitize_content_core.cjs | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/actions/setup/js/sanitize_content_core.cjs b/actions/setup/js/sanitize_content_core.cjs
index bdc1c613705..5ba6bf6a504 100644
--- a/actions/setup/js/sanitize_content_core.cjs
+++ b/actions/setup/js/sanitize_content_core.cjs
@@ -1381,7 +1381,7 @@ function convertXmlTags(s) {
if (malformed) return `(${escapeAngleBrackets(tagContent)})`;
const tagNameMatch = tagContent.match(/^\/?\s*([A-Za-z][A-Za-z0-9]*)/);
- const tagNameDelimiter = tagNameMatch && tagContent[tagNameMatch[0].length];
+ const tagNameDelimiter = tagNameMatch ? tagContent[tagNameMatch[0].length] : undefined;
if (tagNameMatch && (tagNameDelimiter === undefined || /[\s/]/.test(tagNameDelimiter)) && allowedTags.includes(tagNameMatch[1].toLowerCase())) {
return `<${stripDangerousAttributes(tagContent)}>`;
}