diff --git a/.gitea/workflows/comment-pull.yml b/.gitea/workflows/comment-pull.yml
index 70da2f7..e247b8b 100644
--- a/.gitea/workflows/comment-pull.yml
+++ b/.gitea/workflows/comment-pull.yml
@@ -29,6 +29,10 @@ jobs:
         run: |
           docker exec api uv run --no-sync stack bib pull-open-dockets
 
+      - name: Extract attachment text (PDF/DOCX to combined.md)
+        run: |
+          docker exec api uv run --no-sync stack comments extract --workers 4
+
       - name: Index new comments for the chat search
         run: |
           docker exec llm uv run --no-sync stack llm index --collection comments
diff --git a/dev/scripts/backends/gitea.py b/dev/scripts/backends/gitea.py
index 2f2ae86..8c6f9a5 100644
--- a/dev/scripts/backends/gitea.py
+++ b/dev/scripts/backends/gitea.py
@@ -825,7 +825,10 @@ def _gen_comment_pull(runner: str, uv_version: str, **_kw: object) -> tuple[str,
     docket quiet long enough after its comment period seals itself, so the
     job costs nothing between rulemaking seasons. `fetch-pfs-comments` for
     the current year registers a new PFS proposed rule (July) the night it
-    appears. Indexing runs in the llm container (its embeddings host and
+    appears. Attachment text is extracted in the api container (it mounts
+    .state/comments read-write); nothing ran that step on a schedule, so 28
+    dockets were tagged and indexed from "see attached" stubs until
+    2026-11-11. Indexing runs in the llm container (its embeddings host and
     pgvector settings live there). 02:30 runs before the 03:30 notebooks,
     03:40 golden and 04:15 Zotero jobs.
     """
@@ -855,6 +858,10 @@ jobs:
         run: |
           docker exec api uv run --no-sync stack bib pull-open-dockets
 
+      - name: Extract attachment text (PDF/DOCX to combined.md)
+        run: |
+          docker exec api uv run --no-sync stack comments extract --workers 4
+
       - name: Index new comments for the chat search
         run: |
           docker exec llm uv run --no-sync stack llm index --collection comments
diff --git a/src/cli/comments.py b/src/cli/comments.py
index 7c8532f..cfb46ea 100644
--- a/src/cli/comments.py
+++ b/src/cli/comments.py
@@ -122,6 +122,14 @@ def _bib_lookup_factory(use_bib: bool):
     return body
 
 
+def _extraction_complete(docket_dir: Path) -> bool:
+    """Every comment dir under *docket_dir* has its combined.md (a
+    missing dir counts as complete: nothing to extract)."""
+    if not docket_dir.is_dir():
+        return True
+    return all((d / "combined.md").exists() for d in docket_dir.iterdir() if d.is_dir())
+
+
 @app.command()
 def extract(
     root: Path = typer.Option(_DEFAULT_ROOT, "--root", help="Comments root dir."),
@@ -162,8 +170,12 @@ def extract(
 
     lookup, attach_cb, store = _build_bib_helpers(use_bib, attach=use_bib and attach)
     try:
+        # A seal skips finished work only: a docket sealed before anyone
+        # extracted it (28 were, by Nov 2026) is still extracted.
         skip = (
-            set(store.sealed_dockets()) if (store is not None and not force) else set()
+            {d for d in store.sealed_dockets() if _extraction_complete(root / d)}
+            if (store is not None and not force)
+            else set()
         )
         stats = walk_and_extract(
             root,
diff --git a/tests/cli/test_comments_dockets.py b/tests/cli/test_comments_dockets.py
index 6878a90..0af275f 100644
--- a/tests/cli/test_comments_dockets.py
+++ b/tests/cli/test_comments_dockets.py
@@ -112,7 +112,11 @@ def test_dockets_discover_populates_from_tags(mc_connect, mc_client, mc_disc, st
 
 
 @patch("bib.connect")
-def test_extract_skips_sealed_docket(mc_connect, tmp_path: Path, store):
+def test_extract_still_extracts_a_sealed_docket_with_unextracted_dirs(
+    mc_connect, tmp_path: Path, store
+):
+    # 28 dockets were sealed before anyone ran extract, and the seal then
+    # kept them unextracted forever: a seal only skips finished work.
     s = store
     s.docket_upsert(
         Docket(
@@ -125,8 +129,28 @@ def test_extract_skips_sealed_docket(mc_connect, tmp_path: Path, store):
         app, ["extract", "--root", str(tmp_path), "--workers", "1", "--no-attach"]
     )
     assert result.exit_code == 0, result.output
+    assert (cdir / "combined.md").exists()
+
+
+@patch("bib.connect")
+def test_extract_skips_a_fully_extracted_sealed_docket(
+    mc_connect, tmp_path: Path, store
+):
+    s = store
+    s.docket_upsert(
+        Docket(
+            id="CMS-2024-0001", sealed_at="2025-01-01T00:00:00Z", seal_reason="manual"
+        )
+    )
+    mc_connect.return_value = s
+    cdir = _seed_dir(tmp_path, "CMS-2024-0001")
+    (cdir / "combined.md").write_text("done")
+    result = runner.invoke(
+        app, ["extract", "--root", str(tmp_path), "--workers", "1", "--no-attach"]
+    )
+    assert result.exit_code == 0, result.output
     assert "skipped_sealed: 1" in result.output
-    assert not (cdir / "combined.md").exists()
+    assert (cdir / "combined.md").read_text() == "done"
 
 
 @patch("bib.connect")