lint_section_v1
Validate markdown citations in academic writing by checking compliance with citation rules, ensuring proper attribution and formatting for evidence-based content.
Instructions
验证章节引用合规
检查 Agent 写作的 markdown 是否符合引用规则。
Args: pack_id: 证据包 ID markdown: Agent 写作的 markdown 内容 require_citations_per_paragraph: 是否要求每段有引用，默认 False min_citations_per_paragraph: 每段最少引用数，默认 1
Returns: passed, issues[], stats
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| pack_id | Yes | ||
| markdown | Yes | ||
| require_citations_per_paragraph | No | ||
| min_citations_per_paragraph | No |
Output Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
No arguments | |||
Implementation Reference
- src/paperlib_mcp/tools/review.py:677-816 (handler)The main execution logic for the 'lint_section_v1' tool. This function validates the citations in the provided markdown against the chunks in the specified evidence pack (pack_id). It checks for valid chunk existence, pack membership, paragraph citation density (optional), and single-source dominance.
@mcp.tool() def lint_section_v1( pack_id: int, markdown: str, require_citations_per_paragraph: bool = False, min_citations_per_paragraph: int = 1, ) -> dict[str, Any]: """验证章节引用合规 检查 Agent 写作的 markdown 是否符合引用规则。 Args: pack_id: 证据包 ID markdown: Agent 写作的 markdown 内容 require_citations_per_paragraph: 是否要求每段有引用，默认 False min_citations_per_paragraph: 每段最少引用数，默认 1 Returns: passed, issues[], stats """ try: # 获取 pack 中所有 chunk_ids pack_chunks = query_all( "SELECT chunk_id FROM evidence_pack_items WHERE pack_id = %s", (pack_id,), ) if not pack_chunks: return {"error": f"Pack not found or empty: {pack_id}"} valid_chunk_ids = {row["chunk_id"] for row in pack_chunks} # 解析 markdown 中的引用 # 格式: [[chunk:<chunk_id>]] citation_pattern = r"\[\[chunk:(\d+)\]\]" citations = re.findall(citation_pattern, markdown) cited_chunk_ids = [int(c) for c in citations] issues = [] valid_citations = 0 invalid_citations = 0 # 检查每个引用 for chunk_id in cited_chunk_ids: # 检查是否存在 exists = query_one( "SELECT chunk_id FROM chunks WHERE chunk_id = %s", (chunk_id,), ) if not exists: issues.append({ "severity": "error", "rule": "CHUNK_NOT_FOUND", "chunk_id": chunk_id, "message": f"Chunk {chunk_id} does not exist in database", "suggestion": "Remove this citation or use a valid chunk_id from the evidence pack", }) invalid_citations += 1 continue # 检查是否在 pack 内 if chunk_id not in valid_chunk_ids: issues.append({ "severity": "error", "rule": "CHUNK_OUT_OF_PACK", "chunk_id": chunk_id, "message": f"Chunk {chunk_id} is not in evidence pack {pack_id}", "suggestion": "Only cite chunks from the provided evidence pack", }) invalid_citations += 1 continue valid_citations += 1 # 检查段落引用密度（可选） if require_citations_per_paragraph: # 按段落分割 paragraphs = [p.strip() for p in markdown.split("\n\n") if p.strip()] for i, para in enumerate(paragraphs): # 跳过标题行 if para.startswith("#"): continue # 统计该段落的引用 para_citations = re.findall(citation_pattern, para) if len(para_citations) < min_citations_per_paragraph: issues.append({ "severity": "warning", "rule": "LOW_PARAGRAPH_DENSITY", "paragraph_index": i, "message": f"Paragraph {i+1} has {len(para_citations)} citations (minimum: {min_citations_per_paragraph})", "suggestion": f"Add at least {min_citations_per_paragraph - len(para_citations)} more citation(s) to this paragraph", }) # 检查单一来源主导 if cited_chunk_ids: # 获取每个 chunk 的 doc_id chunk_docs = {} for chunk_id in set(cited_chunk_ids): doc = query_one( "SELECT doc_id FROM chunks WHERE chunk_id = %s", (chunk_id,), ) if doc: chunk_docs[chunk_id] = doc["doc_id"] # 统计每个文档被引用次数 doc_cite_counts: dict[str, int] = defaultdict(int) for chunk_id in cited_chunk_ids: doc_id = chunk_docs.get(chunk_id) if doc_id: doc_cite_counts[doc_id] += 1 total = len(cited_chunk_ids) for doc_id, count in doc_cite_counts.items(): if count / total > 0.5: issues.append({ "severity": "warning", "rule": "SINGLE_SOURCE_DOMINANT", "doc_id": doc_id, "message": f"Document {doc_id[:16]}... accounts for {count}/{total} ({count*100//total}%) of citations", "suggestion": "Consider diversifying citations across multiple sources", }) # 判断是否通过 has_errors = any(issue["severity"] == "error" for issue in issues) return { "passed": not has_errors, "issues": issues, "stats": { "total_citations": len(cited_chunk_ids), "valid_citations": valid_citations, "invalid_citations": invalid_citations, "unique_chunks_cited": len(set(cited_chunk_ids)), "pack_chunk_count": len(valid_chunk_ids), "citation_coverage": valid_citations / len(valid_chunk_ids) if valid_chunk_ids else 0, }, } except Exception as e: return {"error": str(e), "passed": False} - src/paperlib_mcp/server.py:47-47 (registration)The call to register_review_tools(mcp) registers all review tools, including 'lint_section_v1', to the FastMCP server instance.
register_review_tools(mcp)