@@ -30,5 +30,7 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.12"
-      # Stdlib only — these tests stub out gh and never call the network.
+      # pyyaml only: the tests stub out gh, never call the network, and the
+      # Anthropic SDK is imported lazily so it isn't needed here.
+      - run: pip install pyyaml
       - run: python -m unittest discover -s pipeline -p "test_*.py" -v
@@ -92,6 +92,14 @@ python pipeline/detect_gaps.py --force                  # report requirement gap
 
 **What it checks (per family, across surfaces):** missing product tab, missing overview / tutorial / how-to, missing hub link, narrative pages sitting in the API Reference tab, ungrouped OpenAPI tags, plus thin pages (< 100 words) and missing frontmatter descriptions.
 
+Pages with children are exempt from the thin-page check: a hub is short because its
+content sits on the child pages, so padding it out is the wrong fix.
+
+Gap `path` values are docs.json nav entries, which carry **no file extension** and can
+share a name with the directory holding their children (`docs/resources/breaking-changes-change-notices`
+is both a page and that folder). Resolve one to a file with `detect_gaps.page_file()`;
+never append to it by hand.
+
 **What it doesn't check:** fine-grained judgment (is *this specific* endpoint missing a how-to, is a page the wrong Diataxis type) — that's the LLM audit (`audit_gaps.py`, follow-up); style/Diataxis prose compliance (the review step); code sample correctness (validation step).
 
 ### 2. Generate Drafts
@@ -353,7 +361,8 @@ The live site structure (which page sits in which tab) is read from `docs.json`
 ## Dependencies
 
 - Python 3.8+
-- `anthropic` (for generate.py, rework.py, review.py, post_review.py)
+- `anthropic` (for generate.py, rework.py, review.py, post_review.py). generate.py
+  imports it lazily, so gap detection, `--dry-run` and the tests work without it
 - `pyyaml` (for all scripts)
 - `gh` CLI (for ship.py, post_review.py, and the open-PR check in generate.py)
 - `mint` / `npx` (optional, for broken-links check in promote.py)
@@ -155,23 +155,44 @@ def ungrouped_tags(endpoints, families):
 # Pages — frontmatter + coarse Diataxis classification                         #
 # --------------------------------------------------------------------------- #
 
-def page_frontmatter(page_path):
-    """Return (frontmatter_dict, raw_content). (None, None) if the file is missing."""
+def page_file(page_path):
+    """The file backing a docs.json page entry, or None if nothing does.
+
+    Nav entries carry no extension, so a nav path is never a usable file path.
+    Worse, a page can share its name with the directory holding its children
+    (docs/resources/breaking-changes-change-notices is both a page and the folder
+    its notices live in), so appending nothing and opening the result hands you a
+    directory. Resolve through here; never build a path from a nav entry by hand.
+    """
     for ext in (".mdx", ".md"):
         fp = REPO_ROOT / (page_path + ext)
-        if fp.exists():
-            try:
-                content = fp.read_text(encoding="utf-8")
-            except Exception:
-                return {}, ""
-            if content.startswith("---"):
-                end = content.find("---", 3)
-                if end != -1:
-                    try:
-                        return (yaml.safe_load(content[3:end]) or {}), content
-                    except yaml.YAMLError:
-                        return {}, content
-            return {}, content
+        if fp.is_file():
+            return fp
+    return None
+
+
+def has_child_pages(all_pages, page_path):
+    """Whether other nav entries sit under this one, making it a parent page."""
+    prefix = page_path + "/"
+    return any(p.startswith(prefix) for p in all_pages)
+
+
+def page_frontmatter(page_path):
+    """Return (frontmatter_dict, raw_content). (None, None) if the file is missing."""
+    fp = page_file(page_path)
+    if fp:
+        try:
+            content = fp.read_text(encoding="utf-8")
+        except OSError:
+            return {}, ""
+        if content.startswith("---"):
+            end = content.find("---", 3)
+            if end != -1:
+                try:
+                    return (yaml.safe_load(content[3:end]) or {}), content
+                except yaml.YAMLError:
+                    return {}, content
+        return {}, content
     return None, None  # page listed in nav but file missing
 
 
@@ -329,14 +350,17 @@ def detect_gaps(docs_json, endpoints, families, section=None, force=False):
 
     # --- Doc-quality checks on non-reference pages ------------------------ #
     if not section:
-        for p in all_doc_pages(docs_json, apiref_tab):
+        doc_pages = all_doc_pages(docs_json, apiref_tab)
+        for p in doc_pages:
             fm, content = page_frontmatter(p)
             if fm is None:
                 continue
             if not fm.get("description"):
                 gaps.append(gap("missing_description", "low", None, path=p,
                                 desc=f"{p} has no frontmatter description"))
-            if len(strip_frontmatter(content).split()) < 100:
+            # A page with children is a hub: it's short because the content sits
+            # on the child pages. Padding it out is the wrong fix, so don't ask.
+            if len(strip_frontmatter(content).split()) < 100 and not has_child_pages(doc_pages, p):
                 gaps.append(gap("thin_page", "medium", None, path=p,
                                 desc=f"{p} has under 100 words"))
 
 
@@ -30,15 +30,9 @@
 
 import yaml
 
-try:
-    import anthropic
-except ImportError:
-    print("Install the Anthropic SDK: pip install anthropic")
-    sys.exit(1)
-
-
 from util import build_authoring_system_prompt
 from open_prs import EXIT_NOTHING_TO_DO, fetch_open_prs, split_claimed_gaps
+from detect_gaps import page_file
 
 REPO_ROOT = Path(__file__).resolve().parent.parent
 OPENAPI_PATH = REPO_ROOT / "api-reference" / "openapi.yaml"
@@ -74,6 +68,25 @@ def load_file(path):
         return ""
 
 
+def load_existing_page(gap):
+    """Content of the page a gap edits, resolved from its nav entry.
+
+    Raises instead of returning nothing. A gap's `path` is a docs.json nav entry
+    with no extension, so opening it directly either fails or (when a directory
+    shares the page's name) hands back a directory. Both used to surface as an
+    empty page, which turned "expand this page" into "write one from scratch"
+    and silently discarded the real content.
+    """
+    page_path = gap.get("path", "")
+    resolved = page_file(page_path)
+    if not resolved:
+        raise FileNotFoundError(f"no .mdx or .md file backs the nav entry '{page_path}'")
+    content = load_file(resolved)
+    if not content.strip():
+        raise ValueError(f"{resolved.relative_to(REPO_ROOT)} is empty, refusing to rewrite it from nothing")
+    return content
+
+
 def load_openapi_for_tags(tags):
     """Extract the portion of the OpenAPI spec whose endpoints carry any of `tags`."""
     with open(OPENAPI_PATH) as f:
@@ -128,14 +141,26 @@ def find_existing_docs_for_family(family_name):
 
     context_docs = {}
     for pp in dict.fromkeys(page_paths):  # dedupe, preserve order
-        for ext in (".mdx", ".md"):
-            full_path = REPO_ROOT / (pp + ext)
-            if full_path.exists():
-                context_docs[pp + ext] = load_file(full_path)
-                break
+        full_path = page_file(pp)
+        if full_path:
+            context_docs[str(full_path.relative_to(REPO_ROOT))] = load_file(full_path)
     return context_docs
 
 
+def make_client():
+    """Build the API client, importing the SDK only when a call is imminent.
+
+    Kept lazy so --dry-run, gap detection and the tests all work without the
+    SDK installed — none of them talk to the model.
+    """
+    try:
+        import anthropic
+    except ImportError:
+        print("Install the Anthropic SDK: pip install anthropic")
+        sys.exit(1)
+    return anthropic.Anthropic()
+
+
 def build_system_prompt():
     # All writing rules live in the agent files, assembled in one place.
     return build_authoring_system_prompt(
@@ -208,7 +233,7 @@ def build_generation_prompt(gap, family_name, openapi_context, existing_docs_con
 
     elif gap_type == "thin_page":
         page_path = gap.get("path", "")
-        page_content = load_file(REPO_ROOT / page_path)
+        page_content = load_existing_page(gap)
         return f"""Expand this thin page. It currently has only {gap.get('word_count', 0)} words.
 
 Current content of {page_path}:
@@ -225,7 +250,7 @@ def build_generation_prompt(gap, family_name, openapi_context, existing_docs_con
 
     elif gap_type == "missing_code_examples":
         page_path = gap.get("path", "")
-        page_content = load_file(REPO_ROOT / page_path)
+        page_content = load_existing_page(gap)
         return f"""Add code examples to this page. It's a guide but has no runnable code.
 
 Current content of {page_path}:
@@ -242,7 +267,7 @@ def build_generation_prompt(gap, family_name, openapi_context, existing_docs_con
 
     elif gap_type == "missing_description":
         page_path = gap.get("path", "")
-        page_content = load_file(REPO_ROOT / page_path)[:500]
+        page_content = load_existing_page(gap)[:500]
         return f"""Generate a frontmatter description for this page.
 
 Current content of {page_path} (first 500 chars):
@@ -292,10 +317,13 @@ def determine_output_path(gap, family_name, content=None):
         slug = slugify(title_from_content(content))
         return DOCS_DIR / family_name.lower() / f"{slug}.mdx"
 
-    elif gap_type in ("thin_page", "missing_code_examples"):
-        return REPO_ROOT / gap["path"]
+    elif gap_type in ("thin_page", "missing_code_examples", "missing_description"):
+        # Resolve the nav entry to the real file. Without the extension the draft
+        # is written as `docs--a--b` and promote.py, which only collects *.mdx,
+        # drops it without a word — the page looks generated but never ships.
+        return page_file(gap.get("path", ""))
 
-    return None  # missing_description handled inline; non-generative types skipped
+    return None  # non-generative types skipped
 
 
 def _predicted_path(gap):
@@ -323,7 +351,9 @@ def _pr_review_hint(claimed):
 
 def apply_description(gap, description):
     """Insert a frontmatter description into an existing page."""
-    page_path = REPO_ROOT / gap["path"]
+    page_path = page_file(gap.get("path", ""))
+    if not page_path:
+        return False
     content = page_path.read_text(encoding="utf-8")
 
     if not content.startswith("---"):
@@ -423,10 +453,14 @@ def main():
                 out_path = determine_output_path(g, family)
                 if out_path:
                     print(f"    -> {out_path.relative_to(REPO_ROOT)}")
+                elif g.get("path"):
+                    print(f"    -> skip (no file backs the nav entry '{g['path']}')")
+                else:
+                    print(f"    -> filename chosen from the generated title")
             print()
         return 0
 
-    client = anthropic.Anthropic()
+    client = make_client()
     system_prompt = build_system_prompt()
 
     run_id = datetime.now(timezone.utc).strftime("%Y%m%d-%H%M%S")
@@ -476,8 +510,12 @@ def main():
                     print(f"  Wrote {draft_path.relative_to(REPO_ROOT)}")
                     generated.append({"gap": gap, "action": "file_written", "path": str(rel), "draft": str(draft_path.relative_to(REPO_ROOT))})
                 else:
-                    print(f"  No output path determined, skipping")
-                    errors.append({"gap": gap, "error": "No output path"})
+                    detail = (
+                        f"no file backs the nav entry '{gap['path']}'"
+                        if gap.get("path") else "no output path could be determined"
+                    )
+                    print(f"  Skipping: {detail}")
+                    errors.append({"gap": gap, "error": detail})
 
         except Exception as e:
             print(f"  Error: {e}")