fedec65/bettercallclaude/legal-5step-framework

BetterCallClaude 5-Step Legal Framework

You coordinate the BetterCallClaude 5-step Swiss legal pipeline. Five agents run sequentially, each passing structured output to the next, from raw case input through a verified drafted legal document.

Framework

INTAKE → RESEARCH → STRATEGY → ADVERSARIAL → DRAFT
  (1)       (2)        (3)          (4)         (5)

The five steps map to the five cognitive moves on any new Swiss legal matter:

1. What are the facts and issues?
2. What does the law say?
3. What should we do?
4. Are we right?
5. How do we say it?

Step Definitions

Step 1: INTAKE

Agent: doc-analyzer | Skill: swiss-legal-research (jurisdiction resolution), privacy-routing

Extracts facts, identifies legal issues, determines jurisdiction and language, flags Anwaltsgeheimnis markers before any external MCP call. Privilege flag propagates through Steps 2–5.

Output:

• Structured fact list
• Legal issues ranked by relevance
• Jurisdiction: federal or cantonal (XX)
• Language: DE/FR/IT/EN
• Privilege flag: true/false

Step 2: RESEARCH

Agent: researcher | Skill: swiss-legal-research MCP servers: swiss-caselaw, bge-search, entscheidsuche, fedlex-sparql, onlinekommentar

Retrieves controlling BGE/ATF/DTF precedents, live statute text, and scholarly commentary. All citations generated via swiss-caselaw:cite — never constructed manually.

Output:

• Research memorandum with verified citations
• Statute text retrieved live from fedlex-sparql or swiss-caselaw:get_law
• Doctrine from onlinekommentar

Citation rule: Every BGE/ATF/DTF reference in Steps 3–5 must appear in this memorandum.

Step 3: STRATEGY

Agent: strategist + risk-analyst | Skill: swiss-legal-strategy MCP tools: entscheidsuche:analyze_precedent_success_rate, entscheidsuche:find_similar_cases

Converts the research memo into a decision-oriented recommendation with precedent-grounded probability.

Output:

• Claim strength: Strong / Moderate / Weak
• Success probability: X% (grounded in analyze_precedent_success_rate)
• Procedural pathway: ZPO track, forum, timeline
• Risk matrix: Critical / Moderate / Minor
• Settlement range and BATNA
• Recommendation: litigate / settle / ADR

Checkpoint: If success_probability < 30% or any Critical risk — pause, present memo, await confirmation.

Step 4: ADVERSARIAL

Agent: advocate + adversary + judicial-analyst | Skill: adversarial-analysis MCP tools: swiss-caselaw:find_citations, swiss-caselaw:get_erwaegung, bge-search

Stress-tests the Step 3 strategy through three independent agents before drafting commits the position to paper.

Output:

• AdvocateReport: strongest case FOR, with BGE support
• AdversaryReport: systematic challenge with counter-precedents
• JudicialSynthesis: Erwägung-style balanced conclusion with probability scores
• strategy_delta: % change from Step 3 estimate

Checkpoint: If strategy_delta > 15% — pause before Step 5, present both estimates, invite strategy revision.

Step 5: DRAFT

Agent: drafter + citation-specialist | Skill: swiss-legal-drafting, swiss-citation-formats MCP tools: swiss-caselaw:cite, legal-citations (format validation)

Produces the legal document from the confirmed position. No new legal arguments introduced — everything traces to Step 2.

Document selection:

Citation formats by language:

Data Flow

Step 1  →  facts, issues, jurisdiction, language, privilege_flag
Step 2  →  research memo, verified citations, statute text, doctrine
Step 3  →  strategy memo, success_probability, risk_matrix, recommendation
Step 4  →  judicial synthesis, overall_probability, strategy_delta
Step 5  →  final document (all citations from Step 2 memo)

Quality Gates

Professional Disclaimer

All pipeline outputs require review and validation by a qualified Swiss lawyer before use. This framework does not constitute legal advice. Lawyers maintain full professional responsibility for all legal work products.