fix: prevent duplicate frontmatter in LLM-generated content

Remove frontmatter format from schema to avoid LLM copying it.
Add strip as fallback in _write_summary and _write_concept create path.

Author: rejojer

openkb/agent/compiler.py

@@ -283,6 +283,10 @@ def _read_concept_briefs(wiki_dir: Path) -> str:
 def _write_summary(wiki_dir: Path, doc_name: str, summary: str,
                     doc_type: str = "short") -> None:
     """Write summary page with frontmatter."""
+    if summary.startswith("---"):
+        end = summary.find("---", 3)
+        if end != -1:
+            summary = summary[end + 3:].lstrip("\n")
     summaries_dir = wiki_dir / "summaries"
     summaries_dir.mkdir(parents=True, exist_ok=True)
     ext = "md" if doc_type == "short" else "json"
@@ -347,6 +351,10 @@ def _write_concept(wiki_dir: Path, name: str, content: str, source_file: str, is
                 existing = fm + body
         path.write_text(existing, encoding="utf-8")
     else:
+        if content.startswith("---"):
+            end = content.find("---", 3)
+            if end != -1:
+                content = content[end + 3:].lstrip("\n")
         fm_lines = [f"sources: [{source_file}]"]
         if brief:
             fm_lines.append(f"brief: {brief}")

openkb/schema.py

@@ -35,10 +35,9 @@
 
 ## Format
 - Use [[wikilink]] to link other wiki pages (e.g., [[concepts/attention]])
-- Summary pages header: `doc_type: short|pageindex` and `full_text: sources/{name}.md|.json`
-- Concept pages header: `sources: [summaries/doc1.md, summaries/doc2.md, ...]`
 - Standard Markdown heading hierarchy
 - Keep each page focused on a single topic
+- Do not include YAML frontmatter (---) in generated content; it is managed by code
 """
 
 # Backward compat alias