Preserve generated GFM checkbox states in reply evidence

Recognize read-only task-list controls from the existing UI renderer and match their visible checked state. Keep raw HTML rejected and add positive and changed-state source/browser regressions.

Co-Authored-By: Paperclip <noreply@paperclip.ing>

tests/runner-e2e/PRODUCT-JOURNEYS.md

@@ -115,7 +115,7 @@ satisfy the check. A pending or suppressed comment can use the same proven nativ
 reply path. A finish summary by itself, an earlier turn, a progress
 bubble, or a cancelled run cannot satisfy it. Private receipts record the run,
 thread anchor and observed text; classic protocol comment matchers stay strict.
-Markdown replies compare complete text and exact link destinations after parsing
+Markdown replies compare complete text, GFM checkbox states and exact link destinations after parsing
 through the UI-owned projection of its existing Markdown renderer and GFM/issue-reference plugins. The observer receives the
 same visible-company prefix list used by the UI, so bare issue keys and URLs
 match their rendered links. Whitespace and inline formatting can change during rendering. A

tests/runner-e2e/native-task-reply.spec.ts

@@ -26,6 +26,8 @@ for (const [source, html, passes] of [
   ["Depends on TREE-604.", "Depends on TREE-604.", true],
   ["Depends on QA-123.", "Depends on <a data-mention-kind='issue' href='/QA/issues/QA-999'>QA-123</a>.", false],
   ["See https://example.com/", "See <a href='https://other.test/'>https://example.com/</a>", false],
+  ["- [x] Finished\n- [ ] Next", "<ul><li><input type='checkbox' checked disabled> Finished</li><li><input type='checkbox' disabled> Next</li></ul>", true],
+  ["- [x] Finished\n- [ ] Next", "<ul><li><input type='checkbox' disabled> Finished</li><li><input type='checkbox' disabled> Next</li></ul>", false],
 ] as const) test(`reloaded auto-linked content: ${source} (${passes})`, async ({ page }) => {
   const url = "https://fixture.test/QA/issues/QA-1";
   await page.route(url, route => route.fulfill({ contentType: "text/html", body:

tests/runner-e2e/native-task-reply.ts

@@ -54,7 +54,7 @@ export function uniqueBoundTaskReply(all: TaskReplyObservation[], run: Row, agen
 
 export function observeTaskReplyElements(elements: Array<HTMLElement | SVGElement>): TaskReplyObservation[] {
   return elements.map(element => {
-    const content: Array<{ text: string } | { href: string; label: string; decorated: boolean }> = [];
+    const content: Array<{ text: string } | { href: string; label: string; decorated: boolean } | { checked: boolean }> = [];
     const nodes: Array<Node | string> = [element];
     while (nodes.length) {
       const node = nodes.pop()!;
@@ -62,6 +62,10 @@ export function observeTaskReplyElements(elements: Array<HTMLElement | SVGElemen
       if (node.nodeType === Node.TEXT_NODE) { content.push({ text: node.textContent ?? "" }); continue; }
       if (!(node instanceof HTMLElement) || node.getAttribute("aria-hidden") === "true"
         || getComputedStyle(node).display === "none" || getComputedStyle(node).visibility === "hidden") continue;
+      if (node instanceof HTMLInputElement && node.type === "checkbox") {
+        if (node.getClientRects().length) content.push({ checked: node.checked });
+        continue;
+      }
       if (node instanceof HTMLAnchorElement) {
         if (node.getClientRects().length && node.innerText.trim()) content.push({ href: node.href, label: node.innerText,
           decorated: node.matches('[data-mention-kind="issue"], [data-sidebar-document-link]') });

tests/runner-e2e/task-reply-content.test.ts

@@ -14,6 +14,12 @@ it("rejects unsupported Markdown rather than stripping unknown or invisible cont
     expect(taskReplyContentMatches({ body: "Visible" }, source)).toBe(false);
 });
 
+it("matches generated GFM checkboxes and rejects changed check states", () => {
+  const observation = { body: "Finished Next", content: [{ checked: true }, { text: " Finished " }, { checked: false }, { text: " Next" }] };
+  expect(taskReplyContentMatches(observation, "- [x] Finished\n- [ ] Next")).toBe(true);
+  expect(taskReplyContentMatches(observation, "- [ ] Finished\n- [ ] Next")).toBe(false);
+});
+
 it("uses production GFM autolinks and issue linking with the viewer's known companies", () => {
   const context = { pageUrl: "https://fixture.test/QA/issues/QA-1", knownCompanyPrefixes: ["QA"] };
   expect(taskReplyContentMatches({ ...context, body: "Depends on QA-123.", content: [{ text: "Depends on " },

tests/runner-e2e/task-reply-content.ts

@@ -2,12 +2,15 @@ import { markdownReplyContent } from "../../ui/src/lib/markdown-reply-content.js
 import { applyCompanyPrefix, extractCompanyPrefixFromPath } from "../../ui/src/lib/company-routes.js";
 import { parseIssueReferenceFromHref } from "../../ui/src/lib/issue-reference.js";
 
-export type ReplyContentPart = { text: string } | { href: string; label?: string; decorated?: boolean };
+export type ReplyContentPart = { text: string } | { href: string; label?: string; decorated?: boolean } | { checked: boolean };
 
 function normalized(parts: ReplyContentPart[], pageUrl: string, source = false) {
   const result: ReplyContentPart[] = [];
   for (const part of parts) {
-    if ("href" in part) {
+    if ("checked" in part) {
+      if (typeof part.checked !== "boolean") return null;
+      result.push({ checked: part.checked });
+    } else if ("href" in part) {
       try {
         const issueRef = source ? parseIssueReferenceFromHref(part.href) : null;
         const href = issueRef ? applyCompanyPrefix(issueRef.href, extractCompanyPrefixFromPath(new URL(pageUrl).pathname)) : part.href;
@@ -37,7 +40,7 @@ export function taskReplyContentMatches(
     // Bind the structured projection to the independently observed innerText.
     // A destination alone cannot attest to arbitrary visible wording.
     if (observation.content.some(part => "href" in part && (typeof part.label !== "string" || !part.label.trim()))) return false;
-    const displayed = observation.content.map(part => "text" in part ? part.text : part.label!).join("").replace(/\s+/g, " ").trim();
+    const displayed = observation.content.map(part => "text" in part ? part.text : "href" in part ? part.label! : "").join("").replace(/\s+/g, " ").trim();
     if (displayed !== observation.body.replace(/\s+/g, " ").trim()) return false;
   }
   const parts = markdownReplyContent(markdown, observation.knownCompanyPrefixes);
@@ -48,6 +51,7 @@ export function taskReplyContentMatches(
   return expected !== null && actual !== null && expected.length === actual.length && expected.every((part, index) => {
     const displayed = actual[index]!;
     return "text" in part ? "text" in displayed && part.text === displayed.text
-      : "href" in displayed && part.href === displayed.href && (part.label === displayed.label || displayed.decorated === true);
+      : "href" in part ? "href" in displayed && part.href === displayed.href && (part.label === displayed.label || displayed.decorated === true)
+        : "checked" in displayed && part.checked === displayed.checked;
   });
 }

ui/src/lib/markdown-reply-content.ts

@@ -3,7 +3,7 @@ import Markdown from "react-markdown";
 import remarkGfm from "remark-gfm";
 import { remarkLinkIssueReferences } from "./issue-reference.js";
 
-export type MarkdownReplyPart = { text: string } | { href: string; label: string };
+export type MarkdownReplyPart = { text: string } | { href: string; label: string } | { checked: boolean };
 type AstNode = { type: string; children?: AstNode[] };
 
 /** The reply observer consumes the UI's existing parser without adding a
@@ -31,8 +31,12 @@ export function markdownReplyContent(markdown: string, knownPrefixes?: string[])
       return true;
     }
     if (Array.isArray(node)) return node.every(child => visit(child, label));
-    if (!isValidElement<{ children?: ReactNode; href?: string }>(node)) return false;
+    if (!isValidElement<{ children?: ReactNode; href?: string; type?: string; checked?: boolean; disabled?: boolean }>(node)) return false;
     if (node.type === Fragment) return visit(node.props.children, label);
+    if (!label && node.type === "input" && node.props.type === "checkbox" && node.props.disabled === true) {
+      parts.push({ checked: node.props.checked === true });
+      return true;
+    }
     if (node.type === "a" && !label && node.props.href) {
       const words: string[] = [];
       if (!visit(node.props.children, words)) return false;