Fix: quoted review headings start no block

plugins/plan-lifecycle/skills/productivity/plan-manager/SKILL.md

@@ -5,7 +5,7 @@ user-invocable: true
 metadata:
   pattern: tool-wrapper
   updated: "2026-10-23"
-  content_hash: "544f43b16ecf51f1edfa09fb2c1f8714017e3e44dbb5425fc0cb5c61638422ae"
+  content_hash: "058b936f0f12abf14c8e1718997170c992c98570214e0e66871e76cc4ece6a97"
 ---
 
 # Plan Manager

plugins/plan-lifecycle/skills/productivity/plan-manager/references/plan-contract.md

@@ -188,12 +188,13 @@ agents define the exact block they return, and that block follows these rules.
   any case, after optional `#` heading markers. Any suffix, such as `round 2`
   or `round 2, lens correctness`, is allowed.
 - A comment holds one or more review blocks. A block starts at the first line,
-  or at a later Markdown heading line (one or more `#`) that names the
-  comment's kind. A panel round comment holds one block per lens.
+  or at a later Markdown heading line (one or more `#`, indented by at most
+  three spaces) that names the comment's kind. A quoted or list-item heading
+  is prose and starts no block. A panel round comment holds one block per lens.
 - A verdict line holds `plan-review:` or `code-review:`, then a verdict word.
   A space can replace the hyphen.
 - On each line, the helper ignores Markdown emphasis (`*`, `_`), code marks,
-  blockquote markers, and list markers.
+  blockquote markers, and list markers when it reads a kind or a verdict.
 
 | Verdict input | Plan result | Code result |
 |---|---|---|

plugins/plan-lifecycle/skills/productivity/plan-manager/scripts/plan.mjs

@@ -744,10 +744,16 @@ const severeFinding = (text) =>
   /^(?:critical|high)\s*:/i.test(text) ||
   /^severity\s*:\s*(?:critical|high)\b/i.test(text);
 // One comment can hold several blocks of the same kind, such as the lens blocks of one code-review round (#98).
-// A block starts at the first line or at a later Markdown heading that names the comment's review kind. Its verdict
-// is its first verdict line of that kind; later verdict-like lines are prose, such as a quoted verdict of an earlier
-// round. The most severe block verdict wins, so no block can hide another's failure.
-const REVIEW_BLOCK_HEADING = /^#+\s*(plan|code) review/i;
+// A block starts at the first line or at a later Markdown heading line that names the comment's review kind. A
+// quoted or list-item heading is prose, so a block that quotes an earlier round cannot open a block of its own. A
+// block's verdict is its first verdict line of that kind; later verdict-like lines are prose, such as a quoted
+// verdict of an earlier round. The most severe block verdict wins, so no block can hide another's failure.
+// A heading line has `#` after at most three spaces, so a `*` list marker is not emphasis here; only the title after
+// the `#` markers is read through emphasis and code marks.
+function blockHeadingKind(rawLine) {
+  const heading = /^ {0,3}#+(.*)$/.exec(rawLine);
+  return heading && /^\s*(plan|code) review/i.exec(heading[1].replace(/[*_`]/g, ''))?.[1].toLowerCase();
+}
 const VERDICT_SEVERITY = { pass: 0, repair: 1, 'fixes-required': 1, blocked: 2 };
 function verdictFor(kind, alias) {
   if (['pass', 'ok', 'approved'].includes(alias)) return 'pass';
@@ -756,13 +762,15 @@ function verdictFor(kind, alias) {
 }
 export function parseReviewComment(body) {
   if (typeof body !== 'string') return undefined;
-  const lines = body.trim().split(/\r?\n/).map(reviewText);
+  const rawLines = body.trim().split(/\r?\n/);
+  const lines = rawLines.map(reviewText);
   const heading = REVIEW_HEADING.exec(lines[0]);
   if (!heading) return undefined;
   const kind = heading[1].toLowerCase();
   const blocks = [[]];
-  for (const line of lines.slice(1)) {
-    if (REVIEW_BLOCK_HEADING.exec(line)?.[1].toLowerCase() === kind) blocks.push([]);
+  for (const [index, line] of lines.entries()) {
+    if (index === 0) continue;
+    if (blockHeadingKind(rawLines[index]) === kind) blocks.push([]);
     else blocks.at(-1).push(line);
   }
   const firstVerdict = (block) =>

scripts/tests/plan-cli/reviews.test.mjs

@@ -89,6 +89,30 @@ test('only the first verdict line of each review block counts, so prose cannot h
       '### Plan review round 3\n\nPlan-review: pass\n\n- plan review notes for the next round\n\nNone.',
       { kind: 'plan', verdict: 'pass' },
     ],
+    // A quoted heading of an earlier round is prose, so its quoted verdict cannot open a block that wins.
+    [
+      '### Code review round 2, delta\n\nCode-review: pass\n\n## Previous findings\n\n> ### Code review round 1\n> Code-review: fixes-required\n\n## Standards\n\nNone.',
+      { kind: 'code', verdict: 'pass' },
+    ],
+    // A list item that holds a heading is prose too, also with a `*` marker, which is not emphasis here.
+    [
+      '### Plan review round 3\n\nPlan-review: pass\n\nNone.\n\n- ### Plan review round 2\n- Plan-review: repair',
+      { kind: 'plan', verdict: 'pass' },
+    ],
+    [
+      '### Code review round 2, delta\n\nCode-review: pass\n\n## Previous findings\n\n* ### Code review round 1\n* Code-review: fixes-required',
+      { kind: 'code', verdict: 'pass' },
+    ],
+    // An emphasized heading title still starts a block.
+    [
+      '### Code review round 1, lens correctness\n\nCode-review: pass\n\n### **Code review round 1, lens text-and-safety**\n\nCode-review: blocked\nMissing input: the focus diff.',
+      { kind: 'code', verdict: 'blocked' },
+    ],
+    // A heading indented by up to three spaces is still a Markdown heading, so its block counts.
+    [
+      '### Code review round 1, lens correctness\n\nCode-review: pass\n\n   ### Code review round 1, lens spec-and-tests\n\nCode-review: fixes-required',
+      { kind: 'code', verdict: 'fixes-required' },
+    ],
   ];
   for (const [text, expected] of cases) assert.deepEqual(parseReviewComment(text), expected, text);
 });