fix(slice-3-review): apply review findings

Must fix:
- frontmatterRewrite: block-style topics with comments preserved
- frontmatterRewrite: CRLF line endings preserved through rewrite
- file_refs: store original_path alongside normalized path for
  case-sensitive filesystems (dead-refs on Linux)

Should fix:
- topics rename/delete: write topics.yaml before rewriting pages
- health broken-links/broken-xwiki: filter archived source pages
- topics list: page_count excludes archived (consistency with topics show)
- topics.yaml header: document comment-stripping behavior
- topics create: ensureFreshIndex before ad-hoc parent lookup
- topics: hoist DB open out of iterating call sites (topicExists helper)
- tag summary: only list newly-added topics

Consider:
- topicExists helper collapsing findTopic + isAdHocTopicInDb
- use indexDbPath consistently in slice-3 commands
- drop impossible error branch in runTopicsDescribe
- fix misleading null/~ comment in topics/yaml.ts

Tests:
- block comments preserved across tag; CRLF round-trip
- parameterized body-byte-preservation (LF/CRLF/mixed, block/flow)
- archived wikilinks and xwiki filter from broken-links
- topics list page_count excludes archived
- topics create --parent ad-hoc (ensureFreshIndex path)
- two-hop cycle via ad-hoc-promotion code path
- --json shape snapshot for topics list/show and health
- normalizePathPreservingCase unit tests

Author: rohans0509

src/commands/health.ts

@@ -226,15 +226,25 @@ function findStale(
  * Only checks active pages — archived pages are allowed to reference
  * files that have since been deleted (that's often why they were
  * archived in the first place).
+ *
+ * We stat the `original_path` (author's casing) rather than the
+ * lowercased `path` — on case-sensitive filesystems like Linux, stat
+ * of a lowercased alias of `src/Dockerfile` returns ENOENT even
+ * though the file exists. macOS and Windows are case-insensitive so
+ * either form resolves there; using the original consistently means
+ * the code behaves identically on every host.
  */
 async function findDeadRefs(
   db: Database.Database,
   scope: HealthScope,
   repoRoot: string,
 ): Promise<{ slug: string; path: string }[]> {
   const rows = db
-    .prepare<[], { slug: string; path: string; is_dir: number }>(
-      `SELECT p.slug, r.path, r.is_dir
+    .prepare<
+      [],
+      { slug: string; path: string; original_path: string; is_dir: number }
+    >(
+      `SELECT p.slug, r.path, r.original_path, r.is_dir
        FROM file_refs r
        JOIN pages p ON p.slug = r.page_slug
        WHERE p.archived_at IS NULL
@@ -244,19 +254,23 @@ async function findDeadRefs(
   const out: { slug: string; path: string }[] = [];
   for (const r of rows) {
     if (!inPageScope(scope, r.slug)) continue;
-    // `r.path` is stored normalized & lowercased. On case-preserving
-    // filesystems (macOS default) this still stats correctly because
-    // HFS+/APFS are case-insensitive. On Linux the repo author is
-    // expected to already be writing canonical casing.
-    const abs = join(repoRoot, r.path);
+    const abs = join(repoRoot, r.original_path);
     if (!existsSync(abs)) {
-      out.push({ slug: r.slug, path: r.path });
+      // Surface the author's casing in the report — matches what's in
+      // the user's frontmatter/wikilink, which is what they'll search
+      // for when fixing the miss.
+      out.push({ slug: r.slug, path: r.original_path });
     }
   }
   return out;
 }
 
-/** Wikilinks whose target slug has no row in `pages`. */
+/**
+ * Wikilinks whose target slug has no row in `pages`. Every other
+ * page-scoped check filters archived source pages out; this one and
+ * `findBrokenXwiki` follow the same rule so the report doesn't flag
+ * broken links from pages that have been retired.
+ */
 function findBrokenLinks(
   db: Database.Database,
   scope: HealthScope,
@@ -265,8 +279,9 @@ function findBrokenLinks(
     .prepare<[], { source_slug: string; target_slug: string }>(
       `SELECT w.source_slug, w.target_slug
        FROM wikilinks w
-       LEFT JOIN pages p ON p.slug = w.target_slug
-       WHERE p.slug IS NULL
+       JOIN pages src ON src.slug = w.source_slug
+       LEFT JOIN pages tgt ON tgt.slug = w.target_slug
+       WHERE tgt.slug IS NULL AND src.archived_at IS NULL
        ORDER BY w.source_slug, w.target_slug`,
     )
     .all();
@@ -289,9 +304,14 @@ async function findBrokenXwiki(
       [],
       { source_slug: string; target_wiki: string; target_slug: string }
     >(
-      `SELECT source_slug, target_wiki, target_slug
-       FROM cross_wiki_links
-       ORDER BY source_slug, target_wiki, target_slug`,
+      // Same archived-source filter as `findBrokenLinks`. Retired pages
+      // shouldn't spam the report with links to wikis that may have
+      // been intentionally retired too.
+      `SELECT x.source_slug, x.target_wiki, x.target_slug
+       FROM cross_wiki_links x
+       JOIN pages src ON src.slug = x.source_slug
+       WHERE src.archived_at IS NULL
+       ORDER BY x.source_slug, x.target_wiki, x.target_slug`,
     )
     .all();
   const out: { source_slug: string; target_wiki: string; target_slug: string }[] = [];

src/commands/info.ts

@@ -147,11 +147,14 @@ function fetchInfo(db: Database.Database, slug: string): InfoRecord | null {
     .map((r) => r.topic_slug);
 
   const refs = db
-    .prepare<[string], { path: string; is_dir: number }>(
-      "SELECT path, is_dir FROM file_refs WHERE page_slug = ? ORDER BY path",
+    .prepare<[string], { original_path: string; is_dir: number }>(
+      // Display the author's casing (`original_path`), not the
+      // lowercased lookup form. The lowercased `path` column is the
+      // query key for `--mentions`; it's not a user-facing string.
+      "SELECT original_path, is_dir FROM file_refs WHERE page_slug = ? ORDER BY original_path",
     )
     .all(slug)
-    .map((r) => ({ path: r.path, is_dir: r.is_dir === 1 }));
+    .map((r) => ({ path: r.original_path, is_dir: r.is_dir === 1 }));
 
   const linksOut = db
     .prepare<[string], { target_slug: string }>(

src/commands/tag.ts

@@ -1,11 +1,9 @@
-import { join } from "node:path";
-
 import { ensureFreshIndex, runIndexer } from "../indexer/index.js";
 import { resolveWikiRoot } from "../indexer/resolveWiki.js";
 import { openIndex } from "../indexer/schema.js";
 import { toKebabCase } from "../slug.js";
 import { rewritePageTopics } from "../topics/frontmatterRewrite.js";
-import { topicsYamlPath } from "../topics/paths.js";
+import { indexDbPath, topicsYamlPath } from "../topics/paths.js";
 import {
   ensureTopic,
   loadTopicsFile,
@@ -92,8 +90,7 @@ export async function runTag(options: TagOptions): Promise<TagCommandOutput> {
   // `tag` — we just need to find each page's file; `ensureFreshIndex`
   // runs first so the common path is consistent.
   await ensureFreshIndex({ repoRoot });
-  const dbPath = join(repoRoot, ".almanac", "index.db");
-  const db = openIndex(dbPath);
+  const db = openIndex(indexDbPath(repoRoot));
 
   // Auto-create missing topics in topics.yaml.
   const yamlPath = topicsYamlPath(repoRoot);
@@ -134,9 +131,15 @@ export async function runTag(options: TagOptions): Promise<TagCommandOutput> {
       });
       if (result.changed) {
         taggedPages += 1;
-        summary.push(`tagged ${page}: ${topics.join(", ")}`);
+        // Only surface the NEWLY ADDED topics — not the full request.
+        // Reporting every requested topic (including ones the page
+        // already had) reads like false positives in commit diffs.
+        const added = result.after.filter((t) => !result.before.includes(t));
+        summary.push(`tagged ${page}: ${added.join(", ")}`);
       } else {
-        summary.push(`no change ${page} (already tagged)`);
+        summary.push(
+          `no change ${page} (already tagged with ${topics.join(", ")})`,
+        );
       }
     }
   } finally {
@@ -180,7 +183,7 @@ export async function runUntag(
   }
 
   await ensureFreshIndex({ repoRoot });
-  const db = openIndex(join(repoRoot, ".almanac", "index.db"));
+  const db = openIndex(indexDbPath(repoRoot));
   let filePath: string;
   try {
     const row = db