correct download labels — show output dir, rename Skipped → Cached

docs/benchmarks.md

@@ -329,7 +329,18 @@ Downloads and extracts the output zip archive for each completed run. Files are
    ├── output files...
 ```
 
-Already-downloaded runs (where the output directory exists) are automatically skipped unless the `-f` / `--force` flag is used, in which case they are overwritten.
+Progress is rendered as a table with one row per run:
+
+```
+Model                File                                     Size       Progress
+──────────────────── ──────────────────────────────────────── ────────── ──────────
+gemini-2.5-pro       gemini-2.5-pro/12345/                    1.24MB     Done
+claude-sonnet-4      claude-sonnet-4/12346/                   2.10MB     Cached
+```
+
+The `Size` column reports the extracted on-disk size of the run's output directory. The `Progress` column is one of `Done` (freshly downloaded), `Cached` (output directory already on disk from a previous download), or `Bad zip` (downloaded archive was corrupt).
+
+Already-downloaded runs (where the output directory exists) are automatically skipped — they appear as `Cached` rows — unless the `-f` / `--force` flag is used, in which case they are overwritten.
 
 When `--include-source` is used, the downloaded zip also contains the kernel session's source files (e.g., `__notebook__.ipynb` and `__notebook_source__.ipynb`).
 

skills/references/benchmarks.md

@@ -254,8 +254,9 @@ kaggle b t download my-task --include-source
 **Behavior details:**
 - Downloads outputs for all runs in a **terminal state** — this includes both `COMPLETED` and `ERRORED` runs (errored runs may still have partial output)
 - Downloads zip archives and extracts them automatically
-- Already-downloaded runs are skipped (use `--force` to re-download): `Skipping gemini-2.5-pro (run 123) — already downloaded to ./my-task/1/gemini-2.5-pro/123`
-- Corrupt zips: Warning printed, raw `.zip` file kept, continues with other models
+- Progress is rendered as a table with `Model | File | Size | Progress` columns. `Size` is the extracted on-disk size of the run's output directory.
+- Already-downloaded runs are skipped (use `--force` to re-download) and shown as a `Cached` row in the table
+- Corrupt zips show as a `Bad zip` row; the raw `.zip` is kept on disk and the next run is processed
 - No downloadable runs (all still in progress): `No downloadable runs yet — N run(s) still in progress. Use 'kaggle b t status my-task' to check progress.`
 - No runs at all: `No runs found for task 'my-task'. Use 'kaggle b t run my-task' to start one.`
 

src/kaggle/api/kaggle_api_extended.py

@@ -7562,9 +7562,7 @@ def benchmarks_tasks_download_cli(self, task, model=None, output=None, include_s
             print(f"Downloading output runs for {task}")
             print(f"Target directory:  {target_dir}/\n")
 
-            display_files = [
-                f"{self._normalize_model_slug(r.model_version_slug)}/{r.id}/{r.id}.zip" for r in downloadable
-            ]
+            display_files = [f"{self._normalize_model_slug(r.model_version_slug)}/{r.id}/" for r in downloadable]
             model_col = max((len(self._normalize_model_slug(r.model_version_slug)) for r in downloadable), default=20)
             model_col = max(model_col, 20)
             file_col = max((len(f) for f in display_files), default=40)
@@ -7575,7 +7573,7 @@ def benchmarks_tasks_download_cli(self, task, model=None, output=None, include_s
             print(f"{'Model':<{model_col}} {'File':<{file_col}} {'Size':<{size_col}} {'Progress':<{prog_col}}")
             print(f"{'─' * model_col} {'─' * file_col} {'─' * size_col} {'─' * prog_col}")
 
-            downloaded, skipped = 0, 0
+            downloaded, cached = 0, 0
             for r, display_file in zip(downloadable, display_files):
                 slug = self._normalize_model_slug(r.model_version_slug)
                 # Hierarchical layout: {output}/{task}/{version}/{model}/{run_id}/
@@ -7584,8 +7582,8 @@ def benchmarks_tasks_download_cli(self, task, model=None, output=None, include_s
 
                 if os.path.isdir(outdir) and not force:
                     size_str = self._format_size(self._dir_size(outdir))
-                    print(f"{row_prefix} {size_str:<{size_col}} {'Skipped':<{prog_col}}")
-                    skipped += 1
+                    print(f"{row_prefix} {size_str:<{size_col}} {'Cached':<{prog_col}}")
+                    cached += 1
                     continue
 
                 dl_request = ApiDownloadBenchmarkTaskRunOutputRequest()
@@ -7605,7 +7603,6 @@ def benchmarks_tasks_download_cli(self, task, model=None, output=None, include_s
                 try:
                     # quiet=True: intermediate zip, extracted and removed below
                     self.download_file(response, zipfile_path, kaggle.http_client(), quiet=True)
-                    size_str = self._format_size(os.path.getsize(zipfile_path)) if os.path.exists(zipfile_path) else ""
                     # Note: extractall() is safe here because the zip originates from
                     # the trusted Kaggle server, not user-supplied input (zip-slip).
                     with zipfile.ZipFile(zipfile_path, "r") as zf:
@@ -7627,11 +7624,12 @@ def benchmarks_tasks_download_cli(self, task, model=None, output=None, include_s
                 if os.path.isdir(outdir):
                     shutil.rmtree(outdir)
                 os.rename(tmp_outdir, outdir)
+                # Report extracted on-disk size, matching the cached branch above.
+                size_str = self._format_size(self._dir_size(outdir))
                 downloaded += 1
                 print(f"{row_prefix} {size_str:<{size_col}} {'Done':<{prog_col}}")
 
-            # Summary
-            parts = [f"{n} run(s) {label}" for n, label in ((downloaded, "downloaded"), (skipped, "skipped")) if n]
+            parts = [f"{n} run(s) {label}" for n, label in ((downloaded, "downloaded"), (cached, "cached")) if n]
             print(f"\nDone: {', '.join(parts) or '0 runs downloaded'}.")
 
     @staticmethod

src/kaggle/test/test_benchmarks_cli.py

@@ -1180,7 +1180,7 @@ def test_download_all_pending_shows_message(self, api, capsys):
         assert "kaggle b t status my-task" in output
 
     def test_download_skips_existing_output(self, api, capsys, tmp_path):
-        """Already-downloaded runs are skipped without making API calls."""
+        """Already-downloaded runs render as Cached without making API calls."""
         _setup_runs_response(api, [_make_run(run_id=42)])
         self._mock_download(api)
         outdir = str(tmp_path / "out")
@@ -1192,13 +1192,16 @@ def test_download_skips_existing_output(self, api, capsys, tmp_path):
 
         output = capsys.readouterr().out
         assert "gemini-pro" in output
-        assert "Skipped" in output
-        assert "1 run(s) skipped" in output
+        assert "Cached" in output
+        assert "1 run(s) cached" in output
+        # The File column shows the output directory, not a .zip path
+        assert "gemini-pro/42/" in output
+        assert "/42/42.zip" not in output
         # No download API call should have been made
         api._mock_benchmarks.download_benchmark_task_run_output.assert_not_called()
 
     def test_download_summary_counts(self, api, capsys, tmp_path):
-        """Download summary shows correct downloaded and skipped counts."""
+        """Download summary shows correct downloaded and cached counts."""
         _setup_runs_response(
             api,
             [_make_run(model="new-model", run_id=1), _make_run(model="old-model", run_id=2)],
@@ -1214,7 +1217,7 @@ def test_download_summary_counts(self, api, capsys, tmp_path):
 
         output = capsys.readouterr().out
         assert "1 run(s) downloaded" in output
-        assert "1 run(s) skipped" in output
+        assert "1 run(s) cached" in output
 
     def test_download_force_overwrites_existing_output(self, api, capsys, tmp_path):
         """Using force=True re-downloads and overwrites existing output."""