markdown-flow 0.2.84__tar.gz → 0.2.85__tar.gz

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: markdown-flow
-Version: 0.2.84
+Version: 0.2.85
 Summary: An agent library designed to parse and process MarkdownFlow documents
 Project-URL: Homepage, https://github.com/ai-shifu/markdown-flow-agent-py
 Project-URL: Bug Tracker, https://github.com/ai-shifu/markdown-flow-agent-py/issues

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/markdown_flow/__init__.py

@@ -51,7 +51,13 @@ Import Guide:
 # Import core classes and enums
 from .core import MarkdownFlow
 from .enums import BlockType, InputType
-from .formatter import ElementType, FormattedElement, StreamFormatter, format_content
+from .formatter import (
+    ElementType,
+    FormattedElement,
+    StreamFormatter,
+    format_content,
+    strip_untagged_code_fences,
+)
 from .llm import LLMProvider, LLMResult, ProcessMode
 from .parser import (
     InteractionParser,
@@ -81,6 +87,7 @@ __all__ = [
     "FormattedElement",
     "StreamFormatter",
     "format_content",
+    "strip_untagged_code_fences",
     # Main utility functions
     "generate_smart_validation_template",
     "extract_interaction_question",
@@ -88,4 +95,4 @@ __all__ = [
     "replace_variables_in_text",
 ]
 
-__version__ = "0.2.84"
+__version__ = "0.2.85"

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/markdown_flow/formatter/__init__.py

@@ -5,6 +5,7 @@ Provides content classification and structured output for SSE streams.
 """
 
 from .format import format_content
+from .sanitize import strip_untagged_code_fences
 from .stream import StreamFormatter
 from .types import ClassifyResult, ElementType, FormattedElement
 
@@ -15,4 +16,5 @@ __all__ = [
     "ClassifyResult",
     "format_content",
     "StreamFormatter",
+    "strip_untagged_code_fences",
 ]

markdown_flow-0.2.85/markdown_flow/formatter/sanitize.py

@@ -0,0 +1,95 @@
+"""
+Content Sanitizer
+
+Utilities for cleaning LLM-generated content before it is classified and
+rendered.
+
+Motivation
+----------
+LLMs intermittently wrap otherwise-renderable output (block-level HTML, SVG
+infographics, plain narrative) in an *untagged* fenced code block, e.g. a bare
+``` line with no info string, sometimes even leaving it unclosed. CommonMark
+faithfully turns that region into a code block, so a renderer shows the raw
+HTML source instead of the intended visual.
+
+This module targets that specific artifact while preserving deliberate code
+blocks. It relies on a simple, explicit convention: **source that is meant to
+be displayed as code carries a language tag** (``` ```html ```, ``` ```python ```,
+``` ```mermaid ```). An *untagged* fence therefore carries no "show the source"
+intent and is treated as noise that can be removed so the wrapped content
+renders normally.
+
+The transformation is opt-in: callers apply it explicitly to generated content.
+It does not change the behaviour of the classifier or ``format_content``.
+"""
+
+from ..parser.code_fence_utils import is_code_fence_end, parse_code_fence_start
+
+
+def _fence_info_string(fence_line: str, fence_char: str) -> str:
+    """Return the info string (language tag) of a fence opening line."""
+    return fence_line.lstrip(" ").lstrip(fence_char).strip()
+
+
+def strip_untagged_code_fences(content: str) -> str:
+    """Remove untagged (language-less) code-fence markers from ``content``.
+
+    An untagged fence is a fence opening line whose info string is empty
+    (a bare ``` or ~~~ with no language). Such markers are dropped so the text
+    they wrapped renders as normal Markdown/HTML instead of code. This handles
+    both paired bare fences and a single unclosed/stray bare fence.
+
+    Language-tagged fences (``` ```html ```, ``` ```mermaid ```, ...) and their
+    contents are left untouched, including their bare closing fence line.
+
+    Args:
+        content: Raw content, typically LLM output.
+
+    Returns:
+        Content with untagged fence markers removed. Everything else, including
+        the lines that were inside an untagged fence, is preserved verbatim.
+
+    Examples:
+        >>> strip_untagged_code_fences("```\\n<div>x</div>\\n```")
+        '<div>x</div>'
+        >>> strip_untagged_code_fences("```html\\n<div>x</div>\\n```")
+        '```html\\n<div>x</div>\\n```'
+    """
+    if not content or ("```" not in content and "~~~" not in content):
+        return content
+
+    lines = content.split("\n")
+    result: list[str] = []
+
+    # When inside a language-tagged fence we copy lines verbatim until the
+    # matching closing fence, so nested bare backticks and the closing fence
+    # itself are preserved.
+    inside_tagged_fence = False
+    open_fence = None
+
+    for line in lines:
+        if inside_tagged_fence:
+            result.append(line)
+            if open_fence is not None and is_code_fence_end(line, open_fence):
+                inside_tagged_fence = False
+                open_fence = None
+            continue
+
+        fence = parse_code_fence_start(line)
+        if fence is None:
+            result.append(line)
+            continue
+
+        info = _fence_info_string(fence.line, fence.char)
+        if info == "":
+            # Untagged fence at the top level: drop the marker line, keep the
+            # content it wrapped. Paired and unclosed bare fences both collapse
+            # here because we never enter a fenced state for them.
+            continue
+
+        # Language-tagged fence: keep it and preserve everything up to its close.
+        result.append(line)
+        inside_tagged_fence = True
+        open_fence = fence
+
+    return "\n".join(result)

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/markdown_flow/system_prompt.md

@@ -138,5 +138,5 @@ Amount of modification < 50% -> Diff; >= 50% -> create a new screen.
 13. Output HTML as plain text. **Do not** use ```html or any code block marker. Do not output like this:
 
 ```html
-<div style="width:100%; min-height:100vh; overflow-x:hidden; overflow-y:auto; display:flex; flex-direction:column; align-items:center; padding:1em; font-size:clamp(12px,calc(100vw/48),3vh)">
+html code here
 ```

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/markdown_flow.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: markdown-flow
-Version: 0.2.84
+Version: 0.2.85
 Summary: An agent library designed to parse and process MarkdownFlow documents
 Project-URL: Homepage, https://github.com/ai-shifu/markdown-flow-agent-py
 Project-URL: Bug Tracker, https://github.com/ai-shifu/markdown-flow-agent-py/issues

{markdown_flow-0.2.84 → markdown_flow-0.2.85}/markdown_flow.egg-info/SOURCES.txt

@@ -20,6 +20,7 @@ markdown_flow/formatter/__init__.py
 markdown_flow/formatter/classifier.py
 markdown_flow/formatter/format.py
 markdown_flow/formatter/patterns.py
+markdown_flow/formatter/sanitize.py
 markdown_flow/formatter/stream.py
 markdown_flow/formatter/types.py
 markdown_flow/parser/__init__.py