Legal Intelligent Search Strategy, Workflow, Tool Mapping, and Output Standards

Problem

Legal answers depend heavily on current statutes, case context, and evidence. If a model answers from training data alone, it may cite outdated provisions, mix public legal sources with private notes, or produce unsupported conclusions in contracts, labor disputes, and litigation. The risk is higher when statutes are revised, judicial interpretations update, or internal company materials must be checked against public law. This skill turns legal Q&A into a repeatable RAG retrieval, tool routing, analysis, and output validation workflow.

How It Works

• Retrieve before reasoning: every legal question first queries Peking University Law MCP, while IMA knowledge base provides parallel supplements from personal files, historical contracts, and subscribed legal updates.
• Route by question type: natural-language queries use ai_pkulaw_search; known statutes use get_article or get_law_item_content; cases, case numbers, and statute recognition map to the corresponding MCP tools.
• Handle fallbacks explicitly: if Peking University Law returns no results, use IMA; if both fail, state that no reliable legal source was found instead of inventing citations. Verify tool names before calls to reduce failures caused by MCP configuration changes.
• Keep analysis dimensions fixed: cover issue extraction, jurisdiction, legal basis, risk, and actionable recommendations, and load Legal skills such as contract-risk-reviewer or legal-concept-deep-dive-mctmilk by task type.

Boundaries and Caveats

The skill targets mainland China legal scenarios and produces AI-assisted analysis, not formal legal advice. Public law sources and internal knowledge bases have different authority, so cite the source type; for major litigation, property division, or high-risk contracts, consult a licensed attorney.