feat: Detect language using pygments

pythonbridge/core/review.py

@@ -1,6 +1,9 @@
 import json
 import logging
 
+from pygments.lexers import get_lexer_for_filename
+from pygments.util import ClassNotFound
+
 from pythonbridge.core.config import load_environment
 from pythonbridge.core.diff_parser import parse_patch, clamp_to_valid
 from pythonbridge.gh.client import get_pr, post_review, create_reaction
@@ -21,6 +24,13 @@ def _severity_emoji(severity: str) -> str:
     return _SEVERITY_EMOJI.get(severity.lower(), "")
 
 
+def _detect_language(filename: str) -> str:
+    try:
+        return get_lexer_for_filename(filename).name
+    except ClassNotFound:
+        return "Unknown"
+
+
 def _build_pr_context(title: str, body: str) -> str:
     return f"\n\n## PR Context\n\n**Title:** {title}\n\n**Description:**\n{body or '_No description provided._'}\n"
 
@@ -68,8 +78,10 @@ def review_pr(payload: dict) -> list[dict]:
 
         # Annotate the diff with real line numbers so the LLM can reference them accurately
         annotated_patch, valid_lines = parse_patch(file.patch)
+        language = _detect_language(file.filename)
+        llm_input = f"Language: {language}\nFile: {file.filename}\n\n{annotated_patch}"
 
-        result = agent_graph.invoke({"pr_input": annotated_patch})
+        result = agent_graph.invoke({"pr_input": llm_input})
         raw_review = result.get("pr_review") if result else None
         comments = _parse_comments(raw_review, file.filename) if raw_review else []
 

pythonbridge/pyproject.toml

@@ -9,6 +9,7 @@ dependencies = [
     "langgraph>=1.0.7",
     "pydantic>=2.12.5",
     "pygithub>=2.8.1",
+    "pygments>=2.20.0",
     "python-dotenv>=1.0.0",
     "tree-sitter>=0.25.2",
     "tree-sitter-python>=0.25.0",