Render Read tool results with pygments via structured payload (closes #170)

claude_code_log/factories/tool_factory.py

@@ -207,8 +207,11 @@ def _parse_cat_n_snippet(
                 system_reminder += line + "\n"
             continue
 
-        # Parse regular code line (format: "  123→content")
-        match = re.match(r"\s+(\d+)→(.*)$", line)
+        # Parse regular code line. Two cat-n separator variants seen in the
+        # wild: the arrow form ("  123→content", used by Edit/Write result
+        # snippets) and the literal-tab form ("775\tcontent", used by Read
+        # results since at least Claude Code 2.1.x — see issue #170).
+        match = re.match(r"\s*(\d+)[\t→](.*)$", line)
         if match:
             line_num = int(match.group(1))
             # Capture the first line number as offset
@@ -255,25 +258,75 @@ def _extract_tool_result_text(tool_result: ToolResultContent) -> str:
 
 
 def parse_read_output(
-    tool_result: ToolResultContent, file_path: Optional[str]
+    tool_result: ToolResultContent,
+    file_path: Optional[str],
+    tool_use_result: Optional[ToolUseResult] = None,
 ) -> Optional[ReadOutput]:
     """Parse Read tool result into structured content.
 
+    Prefers the structured ``toolUseResult.file`` payload (clean content
+    plus accurate ``filePath`` / ``startLine`` / ``numLines`` / ``totalLines``
+    metadata) when available; falls back to re-parsing the ``cat -n``
+    formatted ``tool_result.content`` text for older transcripts.
+
     Args:
-        tool_result: The tool result content
+        tool_result: The tool result content (cat-n formatted text)
         file_path: Path to the file that was read (required for ReadOutput)
+        tool_use_result: Structured toolUseResult; ``file`` dict gives us
+            clean content + exact line metadata without re-parsing cat-n
 
     Returns:
         ReadOutput if parsing succeeds, None otherwise
     """
     if not file_path:
         return None
+
+    # Preferred path: structured toolUseResult.file is byte-clean and carries
+    # the exact line metadata. Avoids the lossy cat-n round-trip and gives us
+    # totalLines (which the text-only fallback cannot recover).
+    if isinstance(tool_use_result, dict):
+        file_info: dict[str, Any] = tool_use_result.get("file") or {}
+        content_field: Any = file_info.get("content")
+        if isinstance(content_field, str):
+            # ``numLines`` may legitimately be 0 for an empty-file read, so
+            # distinguish *absent* from *zero* explicitly — the
+            # ``... or default`` truthiness shortcut would silently promote
+            # the zero case to the fallback. Same for ``totalLines`` and
+            # ``startLine``. ``splitlines()`` over ``split("\n")`` for the
+            # absent-numLines fallback so content ending in ``\n`` does not
+            # tack on a phantom trailing element ("x\ny\n".splitlines() →
+            # ["x", "y"], length 2 not 3).
+            num_lines_raw = file_info.get("numLines")
+            num_lines = (
+                int(num_lines_raw)
+                if num_lines_raw is not None
+                else len(content_field.splitlines())
+            )
+            total_lines_raw = file_info.get("totalLines")
+            total_lines = (
+                int(total_lines_raw) if total_lines_raw is not None else num_lines
+            )
+            start_line_raw = file_info.get("startLine")
+            start_line = int(start_line_raw) if start_line_raw is not None else 1
+            return ReadOutput(
+                file_path=str(file_info.get("filePath") or file_path),
+                content=content_field,
+                start_line=start_line,
+                num_lines=num_lines,
+                total_lines=total_lines,
+                is_truncated=num_lines < total_lines,
+                system_reminder=None,
+            )
+
     if not (content := _extract_tool_result_text(tool_result)):
         return None
 
-    # Check if content matches the cat-n format pattern (line_number → content)
+    # Fallback: parse the cat-n formatted text. Accepts both the tab variant
+    # (Read tool result, Claude Code 2.1.x+) and the arrow variant
+    # (Edit/Write result snippets); see _parse_cat_n_snippet for the
+    # combined regex.
     lines = content.split("\n")
-    if not lines or not re.match(r"\s+\d+→", lines[0]):
+    if not lines or not re.match(r"\s*\d+[\t→]", lines[0]):
         return None
 
     result = _parse_cat_n_snippet(lines)
@@ -288,8 +341,8 @@ def parse_read_output(
         content=code_content,
         start_line=line_offset,
         num_lines=num_lines,
-        total_lines=num_lines,  # We don't know total from result
-        is_truncated=False,  # Can't determine from result
+        total_lines=num_lines,  # Not recoverable from text-only fallback
+        is_truncated=False,  # Same — fallback can't tell
         system_reminder=system_reminder,
     )
 
@@ -1266,6 +1319,7 @@ def parse_taskstop_output(
     "WebFetch",
     "Bash",
     "TaskStop",
+    "Read",
 }
 
 

claude_code_log/html/renderer.py

@@ -830,13 +830,19 @@ def title_WriteInput(self, input: WriteInput, message: TemplateMessage) -> str:
     def title_ReadInput(self, input: ReadInput, message: TemplateMessage) -> str:
         """Title → '📄 Read <file_path>[, lines N-M]'."""
         summary = input.file_path
-        # Add line range info if available
+        # Add line range info if available. ``offset`` in the Read tool's
+        # input is the 1-based starting line number (matches what the
+        # ``toolUseResult.file.startLine`` and the cat-n line numbers in
+        # the rendered content show). ``None`` or ``0`` both mean "start
+        # from line 1". The displayed range is inclusive on both ends, so
+        # the end is ``start + limit - 1`` — not ``start + limit``.
         if input.limit is not None:
-            offset = input.offset or 0
+            start = input.offset if input.offset else 1
+            end = start + input.limit - 1
             if input.limit == 1:
-                summary = f"{summary}, line {offset + 1}"
+                summary = f"{summary}, line {start}"
             else:
-                summary = f"{summary}, lines {offset + 1}-{offset + input.limit}"
+                summary = f"{summary}, lines {start}-{end}"
         return self._tool_title(message, "📄", summary)
 
     def title_GlobInput(self, input: GlobInput, message: TemplateMessage) -> str:

test/test_data/read_tool_pygments.jsonl

@@ -0,0 +1,2 @@
+{"parentUuid":"accdfd56-c3cd-4a7f-8af8-64950d3e0c60","isSidechain":false,"userType":"external","cwd":"/home/cboos/Workspace/github/daain/claude-code-log/carol","sessionId":"5662de58-b7fb-4500-bf91-a60590136031","version":"2.1.132","gitBranch":"HEAD","entrypoint":"cli","requestId":"req_011CatnwLh8V3F79QLJRDHkM","message":{"model":"claude-opus-4-7","id":"msg_01CFwpE2Tz4Mp3mn6SAh5z5G","type":"message","role":"assistant","content":[{"type":"tool_use","id":"toolu_01VQrLRn8EY5jVKGvJMWDVRc","name":"Read","input":{"file_path":"/home/cboos/Workspace/github/daain/claude-code-log/carol/claude_code_log/converter.py","offset":775,"limit":20}}],"stop_reason":"tool_use","stop_sequence":null,"usage":{"input_tokens":0,"cache_creation_input_tokens":1000,"cache_read_input_tokens":1000,"output_tokens":50,"service_tier":"standard"}},"type":"assistant","uuid":"40a71f98-83db-404f-ac30-abb3ed710384","timestamp":"2026-05-10T11:16:00.071Z"}
+{"parentUuid":"40a71f98-83db-404f-ac30-abb3ed710384","isSidechain":false,"userType":"external","cwd":"/home/cboos/Workspace/github/daain/claude-code-log/carol","sessionId":"5662de58-b7fb-4500-bf91-a60590136031","version":"2.1.132","gitBranch":"HEAD","entrypoint":"cli","promptId":"0b653878-b7fb-4c91-a36d-ca8e7fbe790e","type":"user","message":{"role":"user","content":[{"tool_use_id":"toolu_01VQrLRn8EY5jVKGvJMWDVRc","type":"tool_result","content":"775\t    with _dag_warnings_suppressed(silent):\n776\t        tree = build_dag_from_entries(\n777\t            all_messages, sidechain_uuids=unloaded_sidechain_uuids\n778\t        )\n779\t        dag_ordered = traverse_session_tree(tree)\n780\t\n781\t    # Re-add summaries/ai-titles/queue-ops (excluded from DAG since they\n782\t    # lack uuid).\n783\t    non_dag_entries: list[TranscriptEntry] = [\n784\t        e\n785\t        for e in all_messages\n786\t        if isinstance(\n787\t            e,\n788\t            (\n789\t                SummaryTranscriptEntry,\n790\t                AiTitleTranscriptEntry,\n791\t                QueueOperationTranscriptEntry,\n792\t            ),\n793\t        )\n794\t    ]"}]},"uuid":"fe7a0f22-ce99-4a3d-8d4e-9cfac6728071","timestamp":"2026-05-10T11:16:00.196Z","toolUseResult":{"type":"text","file":{"filePath":"/home/cboos/Workspace/github/daain/claude-code-log/carol/claude_code_log/converter.py","content":"    with _dag_warnings_suppressed(silent):\n        tree = build_dag_from_entries(\n            all_messages, sidechain_uuids=unloaded_sidechain_uuids\n        )\n        dag_ordered = traverse_session_tree(tree)\n\n    # Re-add summaries/ai-titles/queue-ops (excluded from DAG since they\n    # lack uuid).\n    non_dag_entries: list[TranscriptEntry] = [\n        e\n        for e in all_messages\n        if isinstance(\n            e,\n            (\n                SummaryTranscriptEntry,\n                AiTitleTranscriptEntry,\n                QueueOperationTranscriptEntry,\n            ),\n        )\n    ]","numLines":20,"startLine":775,"totalLines":2953}},"sourceToolAssistantUUID":"40a71f98-83db-404f-ac30-abb3ed710384"}