apunkt/SurfSense
1
0
Fork 0
mirror of https://github.com/MODSetter/SurfSense.git synced 2026-05-05 05:42:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

refactor: improve Markdown fence handling in report generation and viewer

Browse source
This commit is contained in:
Anish Sarkar 2026-02-19 18:32:50 +05:30
parent f7bbce098b
commit 8ae37bdccf
2 changed files with 16 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

16
surfsense_backend/app/agents/new_chat/tools/report.py
View file

@ -43,6 +43,10 @@ logger = logging.getLogger(__name__)
# Reusable formatting instructions appended to section-level and review prompts. # Reusable formatting instructions appended to section-level and review prompts.
_FORMATTING_RULES = """\ _FORMATTING_RULES = """\
- IMPORTANT: Output raw Markdown directly. Do NOT wrap the entire output in a \
code fence (e.g. ```markdown, ````markdown, or any backtick fence). Individual \
code examples and diagrams inside the report should still use fenced code blocks, \
but the report itself must NOT be enclosed in one.
- Maintain proper Markdown formatting throughout. - Maintain proper Markdown formatting throughout.
- When including code examples, ALWAYS format them as proper fenced code blocks \ - When including code examples, ALWAYS format them as proper fenced code blocks \
with the correct language identifier (e.g. ```java, ```python). Code inside code \ with the correct language identifier (e.g. ```java, ```python). Code inside code \
@ -188,16 +192,20 @@ def _strip_wrapping_code_fences(text: str) -> str:
Handles patterns like: Handles patterns like:
```markdown\\n...content...\\n``` ```markdown\\n...content...\\n```
````markdown\\n...content...\\n````
```md\\n...content...\\n``` ```md\\n...content...\\n```
```\\n...content...\\n``` ```\\n...content...\\n```
```json\\n...content...\\n``` ```json\\n...content...\\n```
Supports 3 or more backticks (LLMs escalate when content has triple-backtick blocks).
""" """
stripped = text.strip() stripped = text.strip()
# Match opening fence with optional language tag # Match opening fence with 3+ backticks and optional language tag
m = re.match(r"^```(?:markdown|md|json)?\s*\n", stripped) m = re.match(r"^(`{3,})(?:markdown|md|json)?\s*\n", stripped)
if m and stripped.endswith("```"): if m:
fence = m.group(1) # e.g. "```" or "````"
if stripped.endswith(fence):
stripped = stripped[m.end() :] # remove opening fence stripped = stripped[m.end() :] # remove opening fence
stripped = stripped[:-3].rstrip() # remove closing fence stripped = stripped[: -len(fence)].rstrip() # remove closing fence
return stripped return stripped

5
surfsense_web/components/markdown-viewer.tsx
View file

@ -24,8 +24,9 @@ interface MarkdownViewerProps {
*/ */
function stripOuterMarkdownFence(content: string): string { function stripOuterMarkdownFence(content: string): string {
const trimmed = content.trim(); const trimmed = content.trim();
const match = trimmed.match(/^```(?:markdown|md)?\s*\n([\s\S]+?)\n```\s*$/); // Match 3+ backtick fences (LLMs escalate to 4+ when content has triple-backtick blocks)
return match ? match[1] : content; const match = trimmed.match(/^(`{3,})(?:markdown|md)?\s*\n([\s\S]+?)\n\1\s*$/);
return match ? match[2] : content;
} }
/** /**