diff --git a/README.md b/README.md
index b856249..909bc1e 100644
--- a/README.md
+++ b/README.md
@@ -124,6 +124,9 @@ ucode configure skills
 # Download mode: fetch every skill in the schema to disk (and register the connection).
 ucode configure skills --location main.default --path /abs/project/dir
 
+# Download a named subset of the schema's skills instead of all of them.
+ucode configure skills --location main.default --skill my-skill
+
 # MCP mode: expose the schema's skills as MCP tools instead of downloading.
 ucode configure skills --location main.default,ml.prod --mcp
 ```
@@ -135,7 +138,10 @@ ucode configure skills --location main.default,ml.prod --mcp
   (plus its bundled files) into both `.claude/skills/` and `.agents/skills/`. `--path` (an existing
   absolute directory) is optional; when omitted, skills are written under your home directory. Any
   pre-existing skill dir prompts before it's overwritten. It then registers a schema-less skills
-  MCP connection, leaving any prior `--mcp` scope untouched.
+  MCP connection, leaving any prior `--mcp` scope untouched. `--skill <name>[,<name>…]` narrows the
+  download to the named skills (by leaf name) from the schema instead of all of them; requested
+  names not found in the schema warn and are skipped. `--skill` requires a single `--location`, is
+  download-only, and is rejected with `--mcp`.
 - **MCP mode** (`--location … --mcp`) sets the connection's location set to exactly `<list>`
   (override-only) and rebuilds its `?schema=` URL; no files are downloaded and `--path` is rejected.
 
@@ -160,6 +166,7 @@ you to run `ucode <agent>` (existing agent sessions need a restart before the MC
 | `ucode configure --agents claude --mcp system.ai.slack` | Configure an agent and register its Databricks MCP server(s) in one command |
 | `ucode configure skills` | Register the skills MCP connection (utility tools only); no skills download |
 | `ucode configure skills --location main.default [--path <dir>]` | Download a schema's skills to disk (under `<dir>`, or your home dir) and register a schema-less skills MCP connection |
+| `ucode configure skills --location main.default --skill my-skill` | Download only the named skill(s) from a schema (comma-separated for several) |
 | `ucode configure skills --location main.default --mcp` | Expose a schema's skills as MCP tools (override-only) instead of downloading |
 
 ## Managed Local Files
diff --git a/src/ucode/cli.py b/src/ucode/cli.py
index 1b42b4b..b7b7a6e 100644
--- a/src/ucode/cli.py
+++ b/src/ucode/cli.py
@@ -1545,6 +1545,15 @@ def configure_skills(
             help="(download) Existing absolute dir to download into; defaults to your home dir.",
         ),
     ] = None,
+    skill: Annotated[
+        str | None,
+        typer.Option(
+            "--skill",
+            help="(download) Download only this comma-separated subset of skills (by leaf "
+            "name, e.g. `my-skill`) from the schema, instead of every skill. Requires a "
+            "single --location; not valid with --mcp.",
+        ),
+    ] = None,
 ) -> None:
     """Configure Databricks Skills for your coding tools.
 
@@ -1554,18 +1563,33 @@ def configure_skills(
     When ``--location`` is provided: with ``--mcp``, sets the connection's scope to
     exactly the listed schemas (no download); otherwise, downloads every skill in
     each schema to disk (under ``--path``, or your home dir when omitted) and
-    registers the MCP connection with utility tools only.
+    registers the MCP connection with utility tools only. ``--skill`` narrows a
+    download to a named subset of a single schema's skills (requires exactly one
+    ``--location``).
     """
     try:
         locations = _parse_skill_locations(location)
+        # `--skill` absent -> None (whole schema); present (even empty) -> the
+        # explicit subset, so `--skill ""` downloads nothing.
+        selected_skills = (
+            None if skill is None else {s.strip() for s in skill.split(",") if s.strip()}
+        )
         if mcp and path is not None:
             raise RuntimeError("--path is not valid with --mcp.")
+        if mcp and selected_skills is not None:
+            raise RuntimeError("--skill is not valid with --mcp; it only applies when downloading.")
         if path is not None and not locations:
             raise RuntimeError("--path only applies when downloading with --location.")
+        if selected_skills is not None and not locations:
+            raise RuntimeError("--skill only applies when downloading with --location.")
+        if selected_skills is not None and len(locations) != 1:
+            raise RuntimeError(
+                f"--skill requires a single --location (got: {', '.join(locations)})."
+            )
         if mcp or not locations:
             configure_skills_mcp_command(locations)
         else:
-            configure_skills_download_command(locations, path=path)
+            configure_skills_download_command(locations, path=path, skills=selected_skills)
     except (RuntimeError, ValueError) as exc:
         print_err(str(exc))
         raise typer.Exit(1) from None
diff --git a/src/ucode/skills_download.py b/src/ucode/skills_download.py
index 8822640..124721e 100644
--- a/src/ucode/skills_download.py
+++ b/src/ucode/skills_download.py
@@ -247,12 +247,21 @@ def _fetch_bundles(
     return results
 
 
-def download_skills(workspace: str, token: str, locations: list[str], path: str | None) -> None:
+def download_skills(
+    workspace: str,
+    token: str,
+    locations: list[str],
+    path: str | None,
+    skills: set[str] | None = None,
+) -> None:
     """Download every skill in each ``<catalog>.<schema>`` location to disk.
 
     Bundles are fetched concurrently (with a progress bar) per schema, then
     written sequentially so overwrite prompts don't interleave. A failure on one
     skill warns and skips it without aborting the batch.
+
+    When ``skills`` is given, only those leaf names are downloaded; names absent
+    from a schema warn and are skipped. ``None`` downloads the whole schema.
     """
     roots = skill_dir_roots(path)
     roots_display = " and ".join(str(root) for root in roots)
@@ -262,6 +271,17 @@ def download_skills(workspace: str, token: str, locations: list[str], path: str
         if reason:
             print_warning(f"Skipping `{location}`: {reason}.")
             continue
+        if skills is not None:
+            unknown = skills - set(leaves)
+            if unknown:
+                print_warning(
+                    f"Skipping requested skill(s) not found in `{location}`: "
+                    f"{', '.join(sorted(unknown))}."
+                )
+            leaves = [leaf for leaf in leaves if leaf in skills]
+            if not leaves:
+                print_note(f"No requested skills to download from `{location}`.")
+                continue
         if not leaves:
             print_note(f"No skills found in `{location}`.")
             continue
@@ -281,17 +301,20 @@ def download_skills(workspace: str, token: str, locations: list[str], path: str
         )
 
 
-def configure_skills_download_command(locations: list[str], *, path: str | None) -> int:
+def configure_skills_download_command(
+    locations: list[str], *, path: str | None, skills: set[str] | None = None
+) -> int:
     """Download every skill in each schema to disk and register the skills connection.
 
     Downloads to ``path`` (or the home dir when None), then registers/keeps the
     schema-less MCP connection. ``skill_locations`` is never touched, so a prior
-    ``--mcp`` set survives a download run."""
+    ``--mcp`` set survives a download run. ``skills`` narrows the download (see
+    ``download_skills``)."""
     state = load_state()
     workspace, profile, clients = setup_mcp_clients(state, "Skills")
     token = get_databricks_token(workspace, profile)
 
-    download_skills(workspace, token, locations, path)
+    download_skills(workspace, token, locations, path, skills)
 
     register_schemaless_skills_connection(state, workspace, profile, clients)
     return 0
diff --git a/tests/test_cli.py b/tests/test_cli.py
index 83ed6df..f5119c2 100644
--- a/tests/test_cli.py
+++ b/tests/test_cli.py
@@ -407,13 +407,63 @@ def test_default_mode_dispatches_download_with_path(self):
                 app, ["configure", "skills", "--location", "a.b", "--path", "/tmp/skills"]
             )
         assert result.exit_code == 0, result.output
-        mock_download.assert_called_once_with(["a.b"], path="/tmp/skills")
+        mock_download.assert_called_once_with(["a.b"], path="/tmp/skills", skills=None)
 
     def test_default_mode_without_path_dispatches_download(self):
         with patch("ucode.cli.configure_skills_download_command") as mock_download:
             result = runner.invoke(app, ["configure", "skills", "--location", "a.b"])
         assert result.exit_code == 0, result.output
-        mock_download.assert_called_once_with(["a.b"], path=None)
+        mock_download.assert_called_once_with(["a.b"], path=None, skills=None)
+
+    def test_skill_filter_dispatches_download_with_subset(self):
+        with patch("ucode.cli.configure_skills_download_command") as mock_download:
+            result = runner.invoke(
+                app, ["configure", "skills", "--location", "a.b", "--skill", "my_skill"]
+            )
+        assert result.exit_code == 0, result.output
+        mock_download.assert_called_once_with(["a.b"], path=None, skills={"my_skill"})
+
+    def test_skill_filter_parses_comma_list(self):
+        with patch("ucode.cli.configure_skills_download_command") as mock_download:
+            result = runner.invoke(
+                app, ["configure", "skills", "--location", "a.b", "--skill", "s1, s2"]
+            )
+        assert result.exit_code == 0, result.output
+        mock_download.assert_called_once_with(["a.b"], path=None, skills={"s1", "s2"})
+
+    def test_skill_with_mcp_exit_1(self):
+        with (
+            patch("ucode.cli.configure_skills_mcp_command") as mock_mcp,
+            patch("ucode.cli.configure_skills_download_command") as mock_download,
+        ):
+            result = runner.invoke(
+                app, ["configure", "skills", "--location", "a.b", "--mcp", "--skill", "my_skill"]
+            )
+        assert result.exit_code == 1
+        assert "--skill" in _strip_ansi(result.output)
+        mock_mcp.assert_not_called()
+        mock_download.assert_not_called()
+
+    def test_skill_without_location_exit_1(self):
+        with (
+            patch("ucode.cli.configure_skills_mcp_command") as mock_mcp,
+            patch("ucode.cli.configure_skills_download_command") as mock_download,
+        ):
+            result = runner.invoke(app, ["configure", "skills", "--skill", "my_skill"])
+        assert result.exit_code == 1
+        assert "--skill" in _strip_ansi(result.output)
+        mock_mcp.assert_not_called()
+        mock_download.assert_not_called()
+
+    def test_skill_with_multiple_locations_exit_1(self):
+        with patch("ucode.cli.configure_skills_download_command") as mock_download:
+            result = runner.invoke(
+                app, ["configure", "skills", "--location", "a.b, c.d", "--skill", "my_skill"]
+            )
+        assert result.exit_code == 1
+        output = _strip_ansi(result.output)
+        assert "--skill requires a single --location" in output
+        mock_download.assert_not_called()
 
     def test_path_with_mcp_exit_1(self):
         with (
diff --git a/tests/test_skills_download.py b/tests/test_skills_download.py
index 33ccf62..bd5e54d 100644
--- a/tests/test_skills_download.py
+++ b/tests/test_skills_download.py
@@ -363,6 +363,60 @@ def test_summary_counts_only_written_skills(self, tmp_path, monkeypatch, capsys)
 
         assert "Downloaded 1/2 skill(s) from `main.default` in" in capsys.readouterr().out
 
+    def test_skill_filter_downloads_only_matching_leaves(self, tmp_path, monkeypatch):
+        monkeypatch.setattr(
+            sd, "list_schema_skills", lambda *a, **k: (["pii-handling", "triage"], None)
+        )
+        monkeypatch.setattr(sd, "fetch_skill_bundle", lambda *a, **k: ({"SKILL.md": b"x"}, None))
+
+        sd.download_skills(WS, "token", ["main.default"], str(tmp_path), {"triage"})
+
+        assert (tmp_path / ".claude/skills/triage/SKILL.md").read_bytes() == b"x"
+        assert not (tmp_path / ".claude/skills/pii-handling").exists()
+
+    def test_skill_filter_warns_on_unknown_and_downloads_rest(self, tmp_path, monkeypatch, capsys):
+        monkeypatch.setattr(sd, "list_schema_skills", lambda *a, **k: (["triage"], None))
+        monkeypatch.setattr(sd, "fetch_skill_bundle", lambda *a, **k: ({"SKILL.md": b"x"}, None))
+
+        sd.download_skills(WS, "token", ["main.default"], str(tmp_path), {"triage", "ghost"})
+
+        out = capsys.readouterr().out
+        assert "Skipping requested skill(s) not found in `main.default`: ghost" in out
+        assert (tmp_path / ".claude/skills/triage/SKILL.md").read_bytes() == b"x"
+
+    def test_empty_skill_filter_downloads_nothing(self, tmp_path, monkeypatch, capsys):
+        monkeypatch.setattr(sd, "list_schema_skills", lambda *a, **k: (["triage"], None))
+        called = []
+        monkeypatch.setattr(
+            sd, "fetch_skill_bundle", lambda *a, **k: called.append(1) or ({"SKILL.md": b"x"}, None)
+        )
+
+        sd.download_skills(WS, "token", ["main.default"], str(tmp_path), set())
+
+        assert called == []
+        assert not (tmp_path / ".claude/skills/triage").exists()
+        # The schema has skills; the filter selected none — distinct from the
+        # empty-schema note.
+        out = capsys.readouterr().out
+        assert "No requested skills to download from `main.default`." in out
+        assert "No skills found" not in out
+
+    def test_empty_schema_reports_no_skills_found(self, tmp_path, monkeypatch, capsys):
+        monkeypatch.setattr(sd, "list_schema_skills", lambda *a, **k: ([], None))
+
+        sd.download_skills(WS, "token", ["main.default"], str(tmp_path), None)
+
+        assert "No skills found in `main.default`." in capsys.readouterr().out
+
+    def test_none_skill_filter_downloads_everything(self, tmp_path, monkeypatch):
+        monkeypatch.setattr(sd, "list_schema_skills", lambda *a, **k: (["a", "b"], None))
+        monkeypatch.setattr(sd, "fetch_skill_bundle", lambda *a, **k: ({"SKILL.md": b"x"}, None))
+
+        sd.download_skills(WS, "token", ["main.default"], str(tmp_path), None)
+
+        assert (tmp_path / ".claude/skills/a/SKILL.md").exists()
+        assert (tmp_path / ".claude/skills/b/SKILL.md").exists()
+
 
 class TestConfigureSkillsDownloadCommand:
     def _stub(self, monkeypatch):
@@ -375,7 +429,9 @@ def _stub(self, monkeypatch):
         monkeypatch.setattr(
             sd,
             "download_skills",
-            lambda ws, tok, locations, path: calls.update(download=(ws, tok, locations, path)),
+            lambda ws, tok, locations, path, skills=None: calls.update(
+                download=(ws, tok, locations, path, skills)
+            ),
         )
         monkeypatch.setattr(
             sd,
@@ -389,7 +445,7 @@ def test_downloads_then_registers_connection(self, monkeypatch):
 
         assert sd.configure_skills_download_command(["a.b"], path="/tmp/skills") == 0
 
-        assert calls["download"] == (WS, "token", ["a.b"], "/tmp/skills")
+        assert calls["download"] == (WS, "token", ["a.b"], "/tmp/skills", None)
         assert calls["register"] == (WS, "profile", ["claude"])
 
     def test_none_path_threads_through(self, monkeypatch):
@@ -397,5 +453,13 @@ def test_none_path_threads_through(self, monkeypatch):
 
         assert sd.configure_skills_download_command(["a.b"], path=None) == 0
 
-        assert calls["download"] == (WS, "token", ["a.b"], None)
+        assert calls["download"] == (WS, "token", ["a.b"], None, None)
+        assert calls["register"] == (WS, "profile", ["claude"])
+
+    def test_skills_filter_threads_through(self, monkeypatch):
+        calls = self._stub(monkeypatch)
+
+        assert sd.configure_skills_download_command(["a.b"], path=None, skills={"triage"}) == 0
+
+        assert calls["download"] == (WS, "token", ["a.b"], None, {"triage"})
         assert calls["register"] == (WS, "profile", ["claude"])