benchflow 0.7.2.dev1998__py3-none-any.whl → 0.7.2.dev2000__py3-none-any.whl

benchflow/cli/traj.py

@@ -292,17 +292,23 @@ def register_traj(app: typer.Typer) -> None:
 def _run_upload(options: _UploadOptions) -> None:
     prompted = options.path is None
     path = options.path or _prompt_for_path()
-    repo_slug: str | None = None
+    detected: _DetectedRepo | None = None
     if options.source_id is not None:
         source_id = options.source_id
     else:
         if options.repo:
-            repo_slug = _detect_repo_slug(path)
-        source_id = f"repo/{repo_slug}" if repo_slug else default_source_id(path)
-    if repo_slug:
-        # Contributor-visible metadata: surface the tag so private-repo
-        # sessions can opt out before anything leaves the machine.
-        console.print(f"Repo: {repo_slug} (use --no-repo to omit)")
+            detected = _detect_repo_slug(path)
+        source_id = f"repo/{detected.slug}" if detected else default_source_id(path)
+    if detected:
+        # Contributor-visible metadata: surface the tag and where it came
+        # from so private-repo sessions can opt out before anything leaves
+        # the machine. The local path stays terminal-only; the uploaded
+        # source id is just repo/<owner>/<name>. Plain print keeps the line
+        # one physical line (no Rich wrapping) so the path stays selectable.
+        print(
+            f"Repo: {detected.slug} "
+            f"(from session cwd {detected.session_cwd}; use --no-repo to omit)"
+        )
     destination = _resolve_destination(options)
 
     with (
@@ -419,25 +425,35 @@ def _infer_email() -> str:
 _SESSION_CWD_SCAN_LINES = 50
 
 
-def _detect_repo_slug(path: Path) -> str | None:
+@dataclass(frozen=True)
+class _DetectedRepo:
+    slug: str
+    session_cwd: Path
+
+
+def _detect_repo_slug(path: Path) -> _DetectedRepo | None:
     """Best-effort ``owner/name`` for the repository the session was about.
 
     Reads the working directory the session recorded (Claude events carry a
-    ``cwd`` field; Codex ``session_meta`` payloads do too), asks that
-    directory's git for the ``origin`` remote, and falls back to the
-    invocation directory. Every failure is silent — repo tagging must never
-    break an upload — and local-path remotes never produce a tag, so no
-    local absolute path can leak into the manifest.
+    ``cwd`` field; Codex ``session_meta`` payloads do too) and asks that
+    directory's git for the ``origin`` remote. The trajectory's own recorded
+    cwd is the ONLY provenance source: there is deliberately no fallback to
+    the upload invocation directory, which mis-attributed sessions recorded
+    outside a repo to whatever checkout the contributor happened to upload
+    from. No session cwd, a missing directory, or no GitHub remote all mean
+    no tag. Every failure is silent — repo tagging must never break an
+    upload — and local-path remotes never produce a tag, so no local
+    absolute path can leak into the manifest.
     """
-    for candidate in (_session_cwd(path), Path.cwd()):
-        if candidate is None or not candidate.is_dir():
-            continue
-        remote = _command_stdout(
-            "git", "-C", str(candidate), "remote", "get-url", "origin"
-        )
-        slug = _repo_slug_from_remote(remote) if remote else None
-        if slug:
-            return slug
+    session_cwd = _session_cwd(path)
+    if session_cwd is None or not session_cwd.is_dir():
+        return None
+    remote = _command_stdout(
+        "git", "-C", str(session_cwd), "remote", "get-url", "origin"
+    )
+    slug = _repo_slug_from_remote(remote) if remote else None
+    if slug:
+        return _DetectedRepo(slug=slug, session_cwd=session_cwd)
     return None
 
 

benchflow/trajectories/viewer.py

@@ -431,26 +431,18 @@ def render_rollout(rollout_dir: Path, prompts: list[str] | None = None) -> str:
     while len(prompts) < len(turn_files):
         prompts.append("")
 
-    first_events = _parse_jsonl(turn_files[0].read_text())
-    sys_event = next((e for e in first_events if e.get("type") == "system"), {})
-    total_cost = 0
-    total_turns_count = 0
-
+    all_events: list[dict] = []
     all_blocks = []
     for i, tf in enumerate(turn_files):
         events = _parse_jsonl(tf.read_text())
+        all_events.extend(events)
         all_blocks.append(render_turn(events, i + 1, prompts[i]))
-        for e in events:
-            if e.get("type") == "result":
-                total_cost += e.get("total_cost_usd", 0)
-                total_turns_count += e.get("num_turns", 0)
-
-    # `or "?"` (not just a .get default): a present-but-null value in the
-    # stream-json (e.g. "session_id": null) bypasses the default and would crash
-    # html.escape() / the [:16] slice below with a raw TypeError.
-    session_id = str(sys_event.get("session_id") or "?")
-    model = str(sys_event.get("model") or "?")
-    version = str(sys_event.get("claude_code_version") or "?")
+
+    badges = _stream_header_badges(all_events)
+    badges.append(("turns", str(len(turn_files))))
+    total_cost = _result_cost(all_events)
+    if total_cost is not None:
+        badges.append(("total cost", f"${total_cost:.4f}"))
 
     return f"""<!DOCTYPE html>
 <html>
@@ -464,14 +456,7 @@ def render_rollout(rollout_dir: Path, prompts: list[str] | None = None) -> str:
 <div class="header">
 {_WORDMARK_HTML}
 <h1>{html.escape(rollout_dir.name)}</h1>
-<div class="meta">
-<span>model: {html.escape(model)}</span>
-<span>session: {html.escape(session_id[:16])}...</span>
-<span>claude code: {html.escape(version)}</span>
-<span>turns: {len(turn_files)}</span>
-<span>total cost: ${total_cost:.4f}</span>
-</div>
-</div>
+{_meta_badges_html(badges)}</div>
 {_join_with_divider(all_blocks)}
 </body>
 </html>"""
@@ -672,18 +657,89 @@ def render_jsonl_file(path: Path) -> str:
     body = render_turn(events, 1, "")
     if not body.strip():
         return _NO_TRAJECTORIES_HTML
-    return _stream_json_page(path.name, events, [body])
+    return _stream_json_page(path.name, events, [body], session_fallback=path.stem)
+
+
+def _stream_header_badges(
+    events: list[dict], *, session_fallback: str | None = None
+) -> list[tuple[str, str]]:
+    """Header badges derivable from what the stream actually contains.
+
+    ``claude -p`` stream-json carries a ``type: system`` init event
+    (``session_id`` / ``model`` / ``claude_code_version``); real ``~/.claude``
+    session files don't — their metadata lives per event (``sessionId``,
+    ``version``) and on assistant events (``message.model``). Pull from
+    whichever is present and omit anything unknown: a header of ``?`` badges
+    at the approve moment reads as a broken viewer, not as missing data.
+    Truthiness (not ``.get`` defaults) also swallows present-but-null values
+    like ``"session_id": null``, which previously needed an explicit guard to
+    avoid a TypeError.
+    """
+    sys_event = next((e for e in events if e.get("type") == "system"), {})
 
+    def _first(values) -> object | None:
+        return next((value for value in values if value), None)
 
-def _stream_json_page(title: str, events: list[dict], turn_blocks: list[str]) -> str:
-    sys_event = next((e for e in events if e.get("type") == "system"), {})
-    total_cost = 0.0
+    model = sys_event.get("model") or _first(
+        event["message"].get("model")
+        for event in events
+        if event.get("type") == "assistant" and isinstance(event.get("message"), dict)
+    )
+    session_id = (
+        sys_event.get("session_id")
+        or _first(event.get("sessionId") for event in events)
+        or session_fallback
+    )
+    version = sys_event.get("claude_code_version") or _first(
+        event.get("version") for event in events
+    )
+
+    badges: list[tuple[str, str]] = []
+    if model:
+        badges.append(("model", str(model)))
+    if session_id:
+        sid = str(session_id)
+        badges.append(("session", sid[:16] + ("..." if len(sid) > 16 else "")))
+    if version:
+        badges.append(("claude code", str(version)))
+    return badges
+
+
+def _result_cost(events: list[dict]) -> float | None:
+    """Total cost summed from result events, or ``None`` when no event
+    carries cost data — so the header can hide the badge instead of
+    asserting a fictional ``$0.0000``."""
+    total = 0.0
+    seen = False
     for event in events:
-        if event.get("type") == "result":
-            total_cost += float(event.get("total_cost_usd") or 0)
-    session_id = str(sys_event.get("session_id") or "?")
-    model = str(sys_event.get("model") or "?")
-    version = str(sys_event.get("claude_code_version") or "?")
+        if event.get("type") == "result" and event.get("total_cost_usd") is not None:
+            total += float(event.get("total_cost_usd") or 0)
+            seen = True
+    return total if seen else None
+
+
+def _meta_badges_html(badges: list[tuple[str, str]]) -> str:
+    """The header's ``.meta`` badge row; empty string when nothing is known."""
+    if not badges:
+        return ""
+    spans = "\n".join(
+        f"<span>{html.escape(label)}: {html.escape(value)}</span>"
+        for label, value in badges
+    )
+    return f'<div class="meta">\n{spans}\n</div>\n'
+
+
+def _stream_json_page(
+    title: str,
+    events: list[dict],
+    turn_blocks: list[str],
+    *,
+    session_fallback: str | None = None,
+) -> str:
+    badges = _stream_header_badges(events, session_fallback=session_fallback)
+    total_cost = _result_cost(events)
+    if total_cost is not None:
+        badges.append(("total cost", f"${total_cost:.4f}"))
     return f"""<!DOCTYPE html>
 <html>
 <head>
@@ -696,13 +752,7 @@ def _stream_json_page(title: str, events: list[dict], turn_blocks: list[str]) ->
 <div class="header">
 {_WORDMARK_HTML}
 <h1>{html.escape(title)}</h1>
-<div class="meta">
-<span>model: {html.escape(model)}</span>
-<span>session: {html.escape(session_id[:16])}...</span>
-<span>claude code: {html.escape(version)}</span>
-<span>total cost: ${total_cost:.4f}</span>
-</div>
-</div>
+{_meta_badges_html(badges)}</div>
 {_join_with_divider(turn_blocks)}
 </body>
 </html>"""

{benchflow-0.7.2.dev1998.dist-info → benchflow-0.7.2.dev2000.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: benchflow
-Version: 0.7.2.dev1998
+Version: 0.7.2.dev2000
 Summary: Multi-turn agent benchmarking with ACP — run any agent, any model, any provider.
 Project-URL: Homepage, https://github.com/benchflow-ai/benchflow
 Project-URL: Repository, https://github.com/benchflow-ai/benchflow

{benchflow-0.7.2.dev1998.dist-info → benchflow-0.7.2.dev2000.dist-info}/RECORD

@@ -116,7 +116,7 @@ benchflow/cli/skills.py,sha256=2Apv1iqvmstWEhY6O-PRdxKc8RYwR6f7MvbYfl2dWm0,7497
 benchflow/cli/tasks.py,sha256=b1XP1EZ79ZtK_1shF65qZyv27EcKKFZYV35M4DzxYKI,18534
 benchflow/cli/trace_import.py,sha256=NHDieFdiKQ-2Sr1-GF8Z5EDcbGsz2qq_kFE0rXacC2o,14705
 benchflow/cli/train.py,sha256=MIP_JxptitH348bWWURpcWt-TmDAw1s8d-RUU7jBaIg,23635
-benchflow/cli/traj.py,sha256=NHglRnCEmckvpVzU6ZqVXOP-3dLt7XjmroGHpBxOQuc,24798
+benchflow/cli/traj.py,sha256=uisYWo-f_R_e06sKJPMObYbvDAPJac0zzyU4byGHU7U,25515
 benchflow/continue_run/__init__.py,sha256=dWidng7kJphVUtbP1a1Ma0Lyjxj5yLgQZIApaKCJUew,1479
 benchflow/continue_run/batch.py,sha256=8BNSInyp_Ol6Fcrg4ezBJdi3cBS12nC8mf963_x4fIg,4146
 benchflow/continue_run/orchestrator.py,sha256=wYoseBNE_rZ0EjNwWDyqfave54xNRe43sCd_O4VJjyk,27676
@@ -291,9 +291,9 @@ benchflow/trajectories/sessions.py,sha256=u_6x1WyT8YtKUVOb6b3myTCHhqn58a8NHCLAvQ
 benchflow/trajectories/tree.py,sha256=c0jyoP9OurDQq-zRtPc66edzWUvLGwQQz0jgtyqSw2U,5576
 benchflow/trajectories/trl_sft_tokenization.py,sha256=-Ef0bP9cuzfzOz5nIv2mEF3taQIEw68h6SM1YO-Mwn4,8643
 benchflow/trajectories/types.py,sha256=EG_TDTh0JveXk5WJX4YMcEDBjfe3526lC7JjDvQmiw8,30067
-benchflow/trajectories/viewer.py,sha256=9TW2uct0W8tWZPdRQsTm6hyunaWY9JQKatYIn1oJo7s,36029
-benchflow-0.7.2.dev1998.dist-info/METADATA,sha256=NGO_FL32422wl7_CwHCaI9twkfulYT2Fyqix8nKcIHI,15565
-benchflow-0.7.2.dev1998.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
-benchflow-0.7.2.dev1998.dist-info/entry_points.txt,sha256=qPMBVtH4qz3P4nMALEAdTbvpxvsVch9RhqshZKQn1dE,84
-benchflow-0.7.2.dev1998.dist-info/licenses/LICENSE,sha256=ohnbHkuBylrTtOvaZJk9Hz7dRwmaEEGAb5ZcBV4Xuhw,10779
-benchflow-0.7.2.dev1998.dist-info/RECORD,,
+benchflow/trajectories/viewer.py,sha256=L-tMiR15qdGBaTlVNgY4dOS1pIrf4PB24MvNic6lj4s,37715
+benchflow-0.7.2.dev2000.dist-info/METADATA,sha256=IYq47CWBBGAdRObAxld55c_Nbt_mma-kzn0JFrADhbw,15565
+benchflow-0.7.2.dev2000.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+benchflow-0.7.2.dev2000.dist-info/entry_points.txt,sha256=qPMBVtH4qz3P4nMALEAdTbvpxvsVch9RhqshZKQn1dE,84
+benchflow-0.7.2.dev2000.dist-info/licenses/LICENSE,sha256=ohnbHkuBylrTtOvaZJk9Hz7dRwmaEEGAb5ZcBV4Xuhw,10779
+benchflow-0.7.2.dev2000.dist-info/RECORD,,