transcript-viewer 0.8.0__tar.gz → 0.9.0__tar.gz

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/PKG-INFO

@@ -1,10 +1,10 @@
 Metadata-Version: 2.5
 Name: transcript-viewer
-Version: 0.8.0
+Version: 0.9.0
 Summary: Browse agent transcripts in a local, dependency-free web viewer.
 License: MIT
 Requires-Python: >=3.12
-Requires-Dist: atif-make>=0.5.0
+Requires-Dist: atif-make>=0.6.0
 Provides-Extra: ai
 Requires-Dist: anthropic>=0.40; extra == 'ai'
 Provides-Extra: parquet

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "transcript-viewer"
-version = "0.8.0"
+version = "0.9.0"
 description = "Browse agent transcripts in a local, dependency-free web viewer."
 readme = "README.md"
 requires-python = ">=3.12"
@@ -10,7 +10,7 @@ license = { text = "MIT" }
 # The floor is not politeness: this viewer imports atif_make.container, which
 # is how a benchmark published as one JSON array of a thousand runs becomes a
 # thousand sessions. An older atif-make fails to import at all.
-dependencies = ["atif-make>=0.5.0"]
+dependencies = ["atif-make>=0.6.0"]
 
 [project.optional-dependencies]
 # Claude-backed explanations are opt-in, so the default install stays

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/src/transcript_viewer/cli.py

@@ -49,7 +49,12 @@ def cmd_view(args: argparse.Namespace) -> int:
         # An explicit path that holds nothing is a mistake worth reporting; an
         # empty library is not, so this only guards the argument.
         if not entries:
-            print(f"transcript-viewer: nothing convertible in {path}", file=sys.stderr)
+            print(
+                f"transcript-viewer: no transcript in {path}. "
+                f"{corpus.describe_contents(path)}, and none of it is a log "
+                f"this can read.",
+                file=sys.stderr,
+            )
             return 1
     else:
         # Only what has been added deliberately. Scanning a machine because the

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/src/transcript_viewer/corpus.py

@@ -9,6 +9,7 @@ something actually asks for that trajectory.
 from __future__ import annotations
 
 import hashlib
+from collections import Counter
 import re
 import json
 from dataclasses import asdict, dataclass
@@ -311,6 +312,7 @@ def scan(roots: list[Path] | None = None, origin: str = "scanned") -> list[Entry
                 sorted(where.rglob("*.jsonl"))
                 + sorted(where.rglob("*.har"))
                 + sorted(where.rglob("*.json"))
+                + sorted(where.rglob("*.md"))
                 # Not a log, but a container of them, opened above.
                 + sorted(where.rglob("*.parquet"))
             )
@@ -367,3 +369,32 @@ def load(path: Path | None = None) -> list[Entry]:
         except TypeError:
             continue
     return entries
+
+
+def describe_contents(root: Path, limit: int = 4000) -> str:
+    """What is inside a path, said in a sentence.
+
+    Reached only when nothing in it converted. "Nothing convertible" on its own
+    invites the reader to doubt the tool rather than the file — one real archive
+    held 1,952 files and an audit that referenced a transcript which had not
+    been shipped with it, and the answer gave no way to tell that from a bug.
+    """
+    if root.is_file():
+        return f"{root.name} ({root.stat().st_size:,} bytes)"
+
+    kinds: Counter[str] = Counter()
+    total = 0
+    for item in root.rglob("*"):
+        if not item.is_file():
+            continue
+        total += 1
+        if total > limit:
+            break
+        kinds[item.suffix.lower() or "no extension"] += 1
+
+    if not total:
+        return "it is empty"
+
+    common = ", ".join(f"{n:,} {ext}" for ext, n in kinds.most_common(4))
+    more = "+" if total > limit else ""
+    return f"{total:,}{more} files inside — {common}"

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/src/transcript_viewer/fetch.py

@@ -58,7 +58,8 @@ HOSTS = {
 # in: a bucket of agent transcripts is far more likely to hold one zip per
 # session than loose JSONL, so excluding archives made S3 useless for exactly
 # the case it was added for.
-LOGS = (".jsonl", ".json", ".har")
+# A transcript is not always JSON: some corpora ship the readable rendering.
+LOGS = (".jsonl", ".json", ".har", ".md", ".markdown")
 # A published dataset is a container of runs rather than a log, but it is still
 # a thing worth fetching: METR's transcripts come as Parquet shards.
 DATASETS = (".parquet",)

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/src/transcript_viewer/page.html

@@ -1728,6 +1728,9 @@ const EXAMPLES=[
   {name:"Redwood agent transcripts",
    url:"s3://rr-agent-transcripts",
    note:"needs an AWS profile set in Settings"},
+  {name:"Redwood sabotage trajectories",
+   url:"s3://rr-strajs",
+   note:"136 transcripts, honest and attack in pairs — needs an AWS profile"},
 ];
 
 function drawExamples(){

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/src/transcript_viewer/viewer.py

@@ -27,6 +27,7 @@ from urllib.parse import parse_qs, unquote, urlparse
 from atif_make.atif import ContentPart, Trajectory
 from . import corpus
 from atif_make.container import is_container, open_container
+from atif_make.detect import FORMATS
 from atif_make.convert import convert
 from transcript_viewer.corpus import Entry, scan
 
@@ -615,9 +616,19 @@ class _Handler(BaseHTTPRequestHandler):
             self._json({"error": str(exc)}, 400)
             return
         if not found:
-            # Nothing usable in it — do not keep the copy around.
+            # Nothing usable in it — do not keep the copy around. Say what was
+            # in it on the way out: an answer of "nothing convertible" with
+            # nothing behind it reads as the tool failing rather than the file
+            # not holding what was hoped.
+            inside = corpus.describe_contents(target)
             shutil.rmtree(home, ignore_errors=True)
-            self._json({"error": f"nothing convertible in {name}"}, 415)
+            self._json(
+                {
+                    "error": f"No transcript in {name}. {inside}, and none of it "
+                    f"is a log this can read ({', '.join(FORMATS)})."
+                },
+                415,
+            )
             return
 
         with self.lock:

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/tests/test_cli.py

@@ -65,7 +65,11 @@ def test_a_path_with_nothing_convertible_reports_rather_than_serving(
 ):
     (tmp_path / "notes.txt").write_text("nothing here")
     assert cli.main([str(tmp_path)]) == 1
-    assert "nothing convertible" in capsys.readouterr().err
+    said = capsys.readouterr().err
+    assert "no transcript" in said
+    # It must also say what it looked at: a bare refusal reads as the tool
+    # failing rather than the path not holding what was hoped for.
+    assert "1 file" in said and ".txt" in said
     assert not served
 
 

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/tests/test_corpus.py

@@ -496,3 +496,36 @@ def test_pieces_unpacked_beside_a_download_are_used_as_they_are(tmp_path):
     entries = corpus.scan([tmp_path])
     assert len(entries) == 3
     assert all(str(beside) in e.path for e in entries)
+
+
+def test_a_path_that_converts_to_nothing_still_says_what_was_in_it(tmp_path):
+    """"Nothing convertible" alone reads as the tool failing, not the file."""
+    (tmp_path / "a").mkdir()
+    (tmp_path / "a" / "notes.md").write_text("# hello")
+    (tmp_path / "a" / "data.json").write_text('{"not": "a transcript"}')
+    (tmp_path / "a" / "run.py").write_text("print(1)")
+
+    said = corpus.describe_contents(tmp_path)
+
+    assert "3 files" in said
+    for ext in (".md", ".json", ".py"):
+        assert ext in said, f"{ext} should be named"
+
+
+def test_describing_a_single_file_names_it(tmp_path):
+    path = tmp_path / "one.jsonl"
+    path.write_text("{}\n")
+    said = corpus.describe_contents(path)
+    assert "one.jsonl" in said and "bytes" in said
+
+
+def test_describing_an_empty_directory_says_so(tmp_path):
+    assert corpus.describe_contents(tmp_path) == "it is empty"
+
+
+def test_describing_a_huge_tree_does_not_read_all_of_it(tmp_path):
+    """A 600 MB archive should not be walked twice to explain itself."""
+    for i in range(60):
+        (tmp_path / f"f{i}.bin").write_text("x")
+    said = corpus.describe_contents(tmp_path, limit=10)
+    assert "+" in said, "it should say the count is a floor, not a total"

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/tests/test_fetch.py

@@ -215,7 +215,11 @@ def test_a_dataset_url_lists_its_convertible_files(monkeypatch):
     )
     assert service == "hf"
     assert label == "owner--name"
-    assert [f.name for f in files] == ["a/transcript.jsonl", "b/metadata.json"]
+    assert [f.name for f in files] == [
+        "a/transcript.jsonl",
+        "a/description.md",
+        "b/metadata.json",
+    ]
     assert all("/resolve/main/" in f.url for f in files)
     assert seen["auth"] == ["tok"], "the token was not sent"
 
@@ -310,11 +314,23 @@ def test_a_blob_url_is_rewritten_to_raw(monkeypatch):
 
 
 def test_nothing_convertible_is_said_plainly(monkeypatch):
-    _stub(monkeypatch, {"api/datasets": [{"type": "file", "path": "readme.md"}]})
+    # Markdown is fetchable — some corpora publish transcripts as nothing else —
+    # so the refusal is tested with something that could not be one.
+    _stub(monkeypatch, {"api/datasets": [{"type": "file", "path": "diagram.png"}]})
     with pytest.raises(fetch.FetchError, match="Nothing convertible"):
         fetch.plan("https://huggingface.co/datasets/owner/name", {})
 
 
+def test_a_transcript_published_as_markdown_is_worth_fetching(monkeypatch):
+    """A bucket of 136 of these was unreachable while .md was filtered out."""
+    _stub(
+        monkeypatch,
+        {"api/datasets": [{"type": "file", "path": "run/transcript_attack.md"}]},
+    )
+    plan = fetch.plan("https://huggingface.co/datasets/owner/name", {})
+    assert [f.name for f in plan.files] == ["run/transcript_attack.md"]
+
+
 def test_too_many_files_is_refused_before_downloading(monkeypatch):
     _stub(
         monkeypatch,
@@ -489,7 +505,7 @@ def test_an_s3_url_lists_what_is_convertible(monkeypatch):
     plan = fetch.plan("s3://rr-agent-transcripts/runs", {"aws": "rw-eng"})
     assert plan.service == "s3"
     assert plan.label == "rr-agent-transcripts"
-    assert [f.name for f in plan.files] == ["a.jsonl", "deep/b.jsonl"]
+    assert [f.name for f in plan.files] == ["a.jsonl", "notes.md", "deep/b.jsonl"]
     assert plan.web == "s3://rr-agent-transcripts/runs"
 
 
@@ -604,7 +620,10 @@ def test_an_archive_counts_as_something_worth_fetching(monkeypatch):
         ]
     }))
     plan = fetch.plan("s3://rr-agent-transcripts/chippy/abc", {"aws": "rw-eng"})
-    assert [f.name for f in plan.files] == ["transcripts.zip"]
+    # The Markdown comes too: a corpus that publishes transcripts as Markdown
+    # and nothing else was unreachable while it was filtered out. One that is
+    # not a transcript simply does not index.
+    assert [f.name for f in plan.files] == ["transcripts.zip", "notes.md"]
 
 
 def test_the_listing_is_capped_rather_than_exhaustive(monkeypatch):
@@ -650,7 +669,8 @@ def test_s3_browsing_returns_folders_and_files(monkeypatch):
     nodes = fetch.browse("s3://bucket", "chippy", {"aws": "rw-eng"})
     assert "--delimiter" in seen["command"]
     assert [(n.kind, n.name) for n in nodes] == [
-        ("folder", "abc"), ("folder", "def"), ("file", "loose.zip"),
+        ("folder", "abc"), ("folder", "def"),
+        ("file", "loose.zip"), ("file", "readme.md"),
     ]
     assert [n.path for n in nodes][0] == "chippy/abc"
 
@@ -667,7 +687,9 @@ def test_github_browsing_reads_one_level(monkeypatch):
         {"type": "file", "path": "logs/notes.md", "size": 3},
     ]})
     nodes = fetch.browse("https://github.com/owner/name/tree/main/logs", "", {})
-    assert [(n.kind, n.name) for n in nodes] == [("folder", "inner"), ("file", "a.jsonl")]
+    assert [(n.kind, n.name) for n in nodes] == [
+        ("folder", "inner"), ("file", "a.jsonl"), ("file", "notes.md"),
+    ]
 
 
 def test_hf_browsing_reads_one_level(monkeypatch):

{transcript_viewer-0.8.0 → transcript_viewer-0.9.0}/uv.lock

@@ -44,11 +44,11 @@ wheels = [
 
 [[package]]
 name = "atif-make"
-version = "0.5.0"
+version = "0.6.0"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/60/2c/9e5d1b1ba278fde5135b8066487e029dd0620161b6a220f5cb2a118a0333/atif_make-0.5.0.tar.gz", hash = "sha256:d4c3122a63b553099f0843d5d1c666136b0bccb1c11e8ebd2b4f9c4a9200cb5e", size = 215842, upload-time = "2026-08-23T05:15:04.439Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/d3/2b/b8cf83efc63bc09a546eb15b4a13a6013413f14b02ce2cc8e491c7884c21/atif_make-0.6.0.tar.gz", hash = "sha256:1aabe7af93cb07cbe23f39dd6f28aab5aa42f96074eb6f0d8312400447493f77", size = 240892, upload-time = "2026-08-24T21:29:42.84Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/00/47/f9a09d10b3d53600c43fa1b22c1d01cb731e628b3572c2e40109de195cfb/atif_make-0.5.0-py3-none-any.whl", hash = "sha256:b52fac5489f5cbc34f947f5d8b21b51a6a04276420d908a96f938e1bf7443791", size = 49769, upload-time = "2026-08-23T05:15:03.265Z" },
+    { url = "https://files.pythonhosted.org/packages/f9/b2/41bc96961d27d0d8794844ff2c716345f4593f6d1c0d7378039d3ce02b19/atif_make-0.6.0-py3-none-any.whl", hash = "sha256:add29b4bfb41682840b2e1bfe1564ac3b97c6eebe3f894c321b30d7d95485373", size = 58826, upload-time = "2026-08-24T21:29:41.798Z" },
 ]
 
 [package.optional-dependencies]
@@ -388,7 +388,7 @@ wheels = [
 
 [[package]]
 name = "transcript-viewer"
-version = "0.8.0"
+version = "0.9.0"
 source = { editable = "." }
 dependencies = [
     { name = "atif-make" },
@@ -410,7 +410,7 @@ dev = [
 [package.metadata]
 requires-dist = [
     { name = "anthropic", marker = "extra == 'ai'", specifier = ">=0.40" },
-    { name = "atif-make", specifier = ">=0.5.0" },
+    { name = "atif-make", specifier = ">=0.6.0" },
     { name = "atif-make", extras = ["parquet"], marker = "extra == 'parquet'", specifier = ">=0.5.0" },
 ]
 provides-extras = ["parquet", "ai"]